├── .gitignore
├── Dockerfile
├── LICENSE
├── NetCap-output
├── benign
│ ├── chrome
│ │ ├── TLSClientHello.csv
│ │ └── TLSServerHello.csv
│ └── firefox
│ │ └── TLSHandshakes.7z
└── malware
│ ├── TLSClientHello.csv
│ └── TLSServerHello.csv
├── README.md
├── anomaly-detect.py
├── feature-reduction-tests
├── features.py
├── graph_data.py
└── reduce_features.py
├── format-data
├── check_ip.py
├── extract_data_csv.py
├── ja3_fingerprints.csv
└── requirements.txt
├── graph
└── SAVE MODEL GRAPHS HERE
├── models
├── ADD TRAINED MODEL HERE
├── ae.h5
├── oc-svm.pkl
├── svm.pkl
└── win-pkl-ver
│ ├── oc-svm.pkl
│ └── svm.pkl
├── requirements.txt
└── test-train-data
└── test_train_data.7z
/.gitignore:
--------------------------------------------------------------------------------
1 | # Cache files
2 | __pycache__/
3 | *.vscode
4 | validate_data_new-all.csv
5 | format-data/TLSClientHello.csv
6 | format-data/TLSServerHello.csv
7 | NetCap-output/benign/validate/TLSClientHello.csv
8 | NetCap-output/benign/validate/TLSServerHello.csv
9 | test-train-data/test_train_data.csv
10 | .ipynb_checkpoints
--------------------------------------------------------------------------------
/Dockerfile:
--------------------------------------------------------------------------------
1 | FROM tensorflow/tensorflow
2 |
3 | COPY requirements.txt /tmp/requirements.txt
4 | RUN pip install -r /tmp/requirements.txt
5 |
6 | RUN mkdir -p /detect/data && \
7 | mkdir -p /detect/models && \
8 | mkdir -p /detect/graph
9 |
10 | COPY ./anomaly-detect.py /detect/anomaly-detect.py
11 | COPY ./test-train-data/test_train_data.csv /detect/data/test_train_data.csv
12 | COPY ./models/* /detect/models/
13 |
14 | ENV RUNNING_IN_DOCKER=True
15 |
16 | WORKDIR /detect
17 |
18 | ENTRYPOINT [ "python3", "/detect/anomaly-detect.py", "--export" ]
19 |
20 | CMD [ "-h" ]
21 |
--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
1 | GNU GENERAL PUBLIC LICENSE
2 | Version 3, 29 June 2007
3 |
4 | Copyright (C) 2007 Free Software Foundation, Inc.
5 | Everyone is permitted to copy and distribute verbatim copies
6 | of this license document, but changing it is not allowed.
7 |
8 | Preamble
9 |
10 | The GNU General Public License is a free, copyleft license for
11 | software and other kinds of works.
12 |
13 | The licenses for most software and other practical works are designed
14 | to take away your freedom to share and change the works. By contrast,
15 | the GNU General Public License is intended to guarantee your freedom to
16 | share and change all versions of a program--to make sure it remains free
17 | software for all its users. We, the Free Software Foundation, use the
18 | GNU General Public License for most of our software; it applies also to
19 | any other work released this way by its authors. You can apply it to
20 | your programs, too.
21 |
22 | When we speak of free software, we are referring to freedom, not
23 | price. Our General Public Licenses are designed to make sure that you
24 | have the freedom to distribute copies of free software (and charge for
25 | them if you wish), that you receive source code or can get it if you
26 | want it, that you can change the software or use pieces of it in new
27 | free programs, and that you know you can do these things.
28 |
29 | To protect your rights, we need to prevent others from denying you
30 | these rights or asking you to surrender the rights. Therefore, you have
31 | certain responsibilities if you distribute copies of the software, or if
32 | you modify it: responsibilities to respect the freedom of others.
33 |
34 | For example, if you distribute copies of such a program, whether
35 | gratis or for a fee, you must pass on to the recipients the same
36 | freedoms that you received. You must make sure that they, too, receive
37 | or can get the source code. And you must show them these terms so they
38 | know their rights.
39 |
40 | Developers that use the GNU GPL protect your rights with two steps:
41 | (1) assert copyright on the software, and (2) offer you this License
42 | giving you legal permission to copy, distribute and/or modify it.
43 |
44 | For the developers' and authors' protection, the GPL clearly explains
45 | that there is no warranty for this free software. For both users' and
46 | authors' sake, the GPL requires that modified versions be marked as
47 | changed, so that their problems will not be attributed erroneously to
48 | authors of previous versions.
49 |
50 | Some devices are designed to deny users access to install or run
51 | modified versions of the software inside them, although the manufacturer
52 | can do so. This is fundamentally incompatible with the aim of
53 | protecting users' freedom to change the software. The systematic
54 | pattern of such abuse occurs in the area of products for individuals to
55 | use, which is precisely where it is most unacceptable. Therefore, we
56 | have designed this version of the GPL to prohibit the practice for those
57 | products. If such problems arise substantially in other domains, we
58 | stand ready to extend this provision to those domains in future versions
59 | of the GPL, as needed to protect the freedom of users.
60 |
61 | Finally, every program is threatened constantly by software patents.
62 | States should not allow patents to restrict development and use of
63 | software on general-purpose computers, but in those that do, we wish to
64 | avoid the special danger that patents applied to a free program could
65 | make it effectively proprietary. To prevent this, the GPL assures that
66 | patents cannot be used to render the program non-free.
67 |
68 | The precise terms and conditions for copying, distribution and
69 | modification follow.
70 |
71 | TERMS AND CONDITIONS
72 |
73 | 0. Definitions.
74 |
75 | "This License" refers to version 3 of the GNU General Public License.
76 |
77 | "Copyright" also means copyright-like laws that apply to other kinds of
78 | works, such as semiconductor masks.
79 |
80 | "The Program" refers to any copyrightable work licensed under this
81 | License. Each licensee is addressed as "you". "Licensees" and
82 | "recipients" may be individuals or organizations.
83 |
84 | To "modify" a work means to copy from or adapt all or part of the work
85 | in a fashion requiring copyright permission, other than the making of an
86 | exact copy. The resulting work is called a "modified version" of the
87 | earlier work or a work "based on" the earlier work.
88 |
89 | A "covered work" means either the unmodified Program or a work based
90 | on the Program.
91 |
92 | To "propagate" a work means to do anything with it that, without
93 | permission, would make you directly or secondarily liable for
94 | infringement under applicable copyright law, except executing it on a
95 | computer or modifying a private copy. Propagation includes copying,
96 | distribution (with or without modification), making available to the
97 | public, and in some countries other activities as well.
98 |
99 | To "convey" a work means any kind of propagation that enables other
100 | parties to make or receive copies. Mere interaction with a user through
101 | a computer network, with no transfer of a copy, is not conveying.
102 |
103 | An interactive user interface displays "Appropriate Legal Notices"
104 | to the extent that it includes a convenient and prominently visible
105 | feature that (1) displays an appropriate copyright notice, and (2)
106 | tells the user that there is no warranty for the work (except to the
107 | extent that warranties are provided), that licensees may convey the
108 | work under this License, and how to view a copy of this License. If
109 | the interface presents a list of user commands or options, such as a
110 | menu, a prominent item in the list meets this criterion.
111 |
112 | 1. Source Code.
113 |
114 | The "source code" for a work means the preferred form of the work
115 | for making modifications to it. "Object code" means any non-source
116 | form of a work.
117 |
118 | A "Standard Interface" means an interface that either is an official
119 | standard defined by a recognized standards body, or, in the case of
120 | interfaces specified for a particular programming language, one that
121 | is widely used among developers working in that language.
122 |
123 | The "System Libraries" of an executable work include anything, other
124 | than the work as a whole, that (a) is included in the normal form of
125 | packaging a Major Component, but which is not part of that Major
126 | Component, and (b) serves only to enable use of the work with that
127 | Major Component, or to implement a Standard Interface for which an
128 | implementation is available to the public in source code form. A
129 | "Major Component", in this context, means a major essential component
130 | (kernel, window system, and so on) of the specific operating system
131 | (if any) on which the executable work runs, or a compiler used to
132 | produce the work, or an object code interpreter used to run it.
133 |
134 | The "Corresponding Source" for a work in object code form means all
135 | the source code needed to generate, install, and (for an executable
136 | work) run the object code and to modify the work, including scripts to
137 | control those activities. However, it does not include the work's
138 | System Libraries, or general-purpose tools or generally available free
139 | programs which are used unmodified in performing those activities but
140 | which are not part of the work. For example, Corresponding Source
141 | includes interface definition files associated with source files for
142 | the work, and the source code for shared libraries and dynamically
143 | linked subprograms that the work is specifically designed to require,
144 | such as by intimate data communication or control flow between those
145 | subprograms and other parts of the work.
146 |
147 | The Corresponding Source need not include anything that users
148 | can regenerate automatically from other parts of the Corresponding
149 | Source.
150 |
151 | The Corresponding Source for a work in source code form is that
152 | same work.
153 |
154 | 2. Basic Permissions.
155 |
156 | All rights granted under this License are granted for the term of
157 | copyright on the Program, and are irrevocable provided the stated
158 | conditions are met. This License explicitly affirms your unlimited
159 | permission to run the unmodified Program. The output from running a
160 | covered work is covered by this License only if the output, given its
161 | content, constitutes a covered work. This License acknowledges your
162 | rights of fair use or other equivalent, as provided by copyright law.
163 |
164 | You may make, run and propagate covered works that you do not
165 | convey, without conditions so long as your license otherwise remains
166 | in force. You may convey covered works to others for the sole purpose
167 | of having them make modifications exclusively for you, or provide you
168 | with facilities for running those works, provided that you comply with
169 | the terms of this License in conveying all material for which you do
170 | not control copyright. Those thus making or running the covered works
171 | for you must do so exclusively on your behalf, under your direction
172 | and control, on terms that prohibit them from making any copies of
173 | your copyrighted material outside their relationship with you.
174 |
175 | Conveying under any other circumstances is permitted solely under
176 | the conditions stated below. Sublicensing is not allowed; section 10
177 | makes it unnecessary.
178 |
179 | 3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180 |
181 | No covered work shall be deemed part of an effective technological
182 | measure under any applicable law fulfilling obligations under article
183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184 | similar laws prohibiting or restricting circumvention of such
185 | measures.
186 |
187 | When you convey a covered work, you waive any legal power to forbid
188 | circumvention of technological measures to the extent such circumvention
189 | is effected by exercising rights under this License with respect to
190 | the covered work, and you disclaim any intention to limit operation or
191 | modification of the work as a means of enforcing, against the work's
192 | users, your or third parties' legal rights to forbid circumvention of
193 | technological measures.
194 |
195 | 4. Conveying Verbatim Copies.
196 |
197 | You may convey verbatim copies of the Program's source code as you
198 | receive it, in any medium, provided that you conspicuously and
199 | appropriately publish on each copy an appropriate copyright notice;
200 | keep intact all notices stating that this License and any
201 | non-permissive terms added in accord with section 7 apply to the code;
202 | keep intact all notices of the absence of any warranty; and give all
203 | recipients a copy of this License along with the Program.
204 |
205 | You may charge any price or no price for each copy that you convey,
206 | and you may offer support or warranty protection for a fee.
207 |
208 | 5. Conveying Modified Source Versions.
209 |
210 | You may convey a work based on the Program, or the modifications to
211 | produce it from the Program, in the form of source code under the
212 | terms of section 4, provided that you also meet all of these conditions:
213 |
214 | a) The work must carry prominent notices stating that you modified
215 | it, and giving a relevant date.
216 |
217 | b) The work must carry prominent notices stating that it is
218 | released under this License and any conditions added under section
219 | 7. This requirement modifies the requirement in section 4 to
220 | "keep intact all notices".
221 |
222 | c) You must license the entire work, as a whole, under this
223 | License to anyone who comes into possession of a copy. This
224 | License will therefore apply, along with any applicable section 7
225 | additional terms, to the whole of the work, and all its parts,
226 | regardless of how they are packaged. This License gives no
227 | permission to license the work in any other way, but it does not
228 | invalidate such permission if you have separately received it.
229 |
230 | d) If the work has interactive user interfaces, each must display
231 | Appropriate Legal Notices; however, if the Program has interactive
232 | interfaces that do not display Appropriate Legal Notices, your
233 | work need not make them do so.
234 |
235 | A compilation of a covered work with other separate and independent
236 | works, which are not by their nature extensions of the covered work,
237 | and which are not combined with it such as to form a larger program,
238 | in or on a volume of a storage or distribution medium, is called an
239 | "aggregate" if the compilation and its resulting copyright are not
240 | used to limit the access or legal rights of the compilation's users
241 | beyond what the individual works permit. Inclusion of a covered work
242 | in an aggregate does not cause this License to apply to the other
243 | parts of the aggregate.
244 |
245 | 6. Conveying Non-Source Forms.
246 |
247 | You may convey a covered work in object code form under the terms
248 | of sections 4 and 5, provided that you also convey the
249 | machine-readable Corresponding Source under the terms of this License,
250 | in one of these ways:
251 |
252 | a) Convey the object code in, or embodied in, a physical product
253 | (including a physical distribution medium), accompanied by the
254 | Corresponding Source fixed on a durable physical medium
255 | customarily used for software interchange.
256 |
257 | b) Convey the object code in, or embodied in, a physical product
258 | (including a physical distribution medium), accompanied by a
259 | written offer, valid for at least three years and valid for as
260 | long as you offer spare parts or customer support for that product
261 | model, to give anyone who possesses the object code either (1) a
262 | copy of the Corresponding Source for all the software in the
263 | product that is covered by this License, on a durable physical
264 | medium customarily used for software interchange, for a price no
265 | more than your reasonable cost of physically performing this
266 | conveying of source, or (2) access to copy the
267 | Corresponding Source from a network server at no charge.
268 |
269 | c) Convey individual copies of the object code with a copy of the
270 | written offer to provide the Corresponding Source. This
271 | alternative is allowed only occasionally and noncommercially, and
272 | only if you received the object code with such an offer, in accord
273 | with subsection 6b.
274 |
275 | d) Convey the object code by offering access from a designated
276 | place (gratis or for a charge), and offer equivalent access to the
277 | Corresponding Source in the same way through the same place at no
278 | further charge. You need not require recipients to copy the
279 | Corresponding Source along with the object code. If the place to
280 | copy the object code is a network server, the Corresponding Source
281 | may be on a different server (operated by you or a third party)
282 | that supports equivalent copying facilities, provided you maintain
283 | clear directions next to the object code saying where to find the
284 | Corresponding Source. Regardless of what server hosts the
285 | Corresponding Source, you remain obligated to ensure that it is
286 | available for as long as needed to satisfy these requirements.
287 |
288 | e) Convey the object code using peer-to-peer transmission, provided
289 | you inform other peers where the object code and Corresponding
290 | Source of the work are being offered to the general public at no
291 | charge under subsection 6d.
292 |
293 | A separable portion of the object code, whose source code is excluded
294 | from the Corresponding Source as a System Library, need not be
295 | included in conveying the object code work.
296 |
297 | A "User Product" is either (1) a "consumer product", which means any
298 | tangible personal property which is normally used for personal, family,
299 | or household purposes, or (2) anything designed or sold for incorporation
300 | into a dwelling. In determining whether a product is a consumer product,
301 | doubtful cases shall be resolved in favor of coverage. For a particular
302 | product received by a particular user, "normally used" refers to a
303 | typical or common use of that class of product, regardless of the status
304 | of the particular user or of the way in which the particular user
305 | actually uses, or expects or is expected to use, the product. A product
306 | is a consumer product regardless of whether the product has substantial
307 | commercial, industrial or non-consumer uses, unless such uses represent
308 | the only significant mode of use of the product.
309 |
310 | "Installation Information" for a User Product means any methods,
311 | procedures, authorization keys, or other information required to install
312 | and execute modified versions of a covered work in that User Product from
313 | a modified version of its Corresponding Source. The information must
314 | suffice to ensure that the continued functioning of the modified object
315 | code is in no case prevented or interfered with solely because
316 | modification has been made.
317 |
318 | If you convey an object code work under this section in, or with, or
319 | specifically for use in, a User Product, and the conveying occurs as
320 | part of a transaction in which the right of possession and use of the
321 | User Product is transferred to the recipient in perpetuity or for a
322 | fixed term (regardless of how the transaction is characterized), the
323 | Corresponding Source conveyed under this section must be accompanied
324 | by the Installation Information. But this requirement does not apply
325 | if neither you nor any third party retains the ability to install
326 | modified object code on the User Product (for example, the work has
327 | been installed in ROM).
328 |
329 | The requirement to provide Installation Information does not include a
330 | requirement to continue to provide support service, warranty, or updates
331 | for a work that has been modified or installed by the recipient, or for
332 | the User Product in which it has been modified or installed. Access to a
333 | network may be denied when the modification itself materially and
334 | adversely affects the operation of the network or violates the rules and
335 | protocols for communication across the network.
336 |
337 | Corresponding Source conveyed, and Installation Information provided,
338 | in accord with this section must be in a format that is publicly
339 | documented (and with an implementation available to the public in
340 | source code form), and must require no special password or key for
341 | unpacking, reading or copying.
342 |
343 | 7. Additional Terms.
344 |
345 | "Additional permissions" are terms that supplement the terms of this
346 | License by making exceptions from one or more of its conditions.
347 | Additional permissions that are applicable to the entire Program shall
348 | be treated as though they were included in this License, to the extent
349 | that they are valid under applicable law. If additional permissions
350 | apply only to part of the Program, that part may be used separately
351 | under those permissions, but the entire Program remains governed by
352 | this License without regard to the additional permissions.
353 |
354 | When you convey a copy of a covered work, you may at your option
355 | remove any additional permissions from that copy, or from any part of
356 | it. (Additional permissions may be written to require their own
357 | removal in certain cases when you modify the work.) You may place
358 | additional permissions on material, added by you to a covered work,
359 | for which you have or can give appropriate copyright permission.
360 |
361 | Notwithstanding any other provision of this License, for material you
362 | add to a covered work, you may (if authorized by the copyright holders of
363 | that material) supplement the terms of this License with terms:
364 |
365 | a) Disclaiming warranty or limiting liability differently from the
366 | terms of sections 15 and 16 of this License; or
367 |
368 | b) Requiring preservation of specified reasonable legal notices or
369 | author attributions in that material or in the Appropriate Legal
370 | Notices displayed by works containing it; or
371 |
372 | c) Prohibiting misrepresentation of the origin of that material, or
373 | requiring that modified versions of such material be marked in
374 | reasonable ways as different from the original version; or
375 |
376 | d) Limiting the use for publicity purposes of names of licensors or
377 | authors of the material; or
378 |
379 | e) Declining to grant rights under trademark law for use of some
380 | trade names, trademarks, or service marks; or
381 |
382 | f) Requiring indemnification of licensors and authors of that
383 | material by anyone who conveys the material (or modified versions of
384 | it) with contractual assumptions of liability to the recipient, for
385 | any liability that these contractual assumptions directly impose on
386 | those licensors and authors.
387 |
388 | All other non-permissive additional terms are considered "further
389 | restrictions" within the meaning of section 10. If the Program as you
390 | received it, or any part of it, contains a notice stating that it is
391 | governed by this License along with a term that is a further
392 | restriction, you may remove that term. If a license document contains
393 | a further restriction but permits relicensing or conveying under this
394 | License, you may add to a covered work material governed by the terms
395 | of that license document, provided that the further restriction does
396 | not survive such relicensing or conveying.
397 |
398 | If you add terms to a covered work in accord with this section, you
399 | must place, in the relevant source files, a statement of the
400 | additional terms that apply to those files, or a notice indicating
401 | where to find the applicable terms.
402 |
403 | Additional terms, permissive or non-permissive, may be stated in the
404 | form of a separately written license, or stated as exceptions;
405 | the above requirements apply either way.
406 |
407 | 8. Termination.
408 |
409 | You may not propagate or modify a covered work except as expressly
410 | provided under this License. Any attempt otherwise to propagate or
411 | modify it is void, and will automatically terminate your rights under
412 | this License (including any patent licenses granted under the third
413 | paragraph of section 11).
414 |
415 | However, if you cease all violation of this License, then your
416 | license from a particular copyright holder is reinstated (a)
417 | provisionally, unless and until the copyright holder explicitly and
418 | finally terminates your license, and (b) permanently, if the copyright
419 | holder fails to notify you of the violation by some reasonable means
420 | prior to 60 days after the cessation.
421 |
422 | Moreover, your license from a particular copyright holder is
423 | reinstated permanently if the copyright holder notifies you of the
424 | violation by some reasonable means, this is the first time you have
425 | received notice of violation of this License (for any work) from that
426 | copyright holder, and you cure the violation prior to 30 days after
427 | your receipt of the notice.
428 |
429 | Termination of your rights under this section does not terminate the
430 | licenses of parties who have received copies or rights from you under
431 | this License. If your rights have been terminated and not permanently
432 | reinstated, you do not qualify to receive new licenses for the same
433 | material under section 10.
434 |
435 | 9. Acceptance Not Required for Having Copies.
436 |
437 | You are not required to accept this License in order to receive or
438 | run a copy of the Program. Ancillary propagation of a covered work
439 | occurring solely as a consequence of using peer-to-peer transmission
440 | to receive a copy likewise does not require acceptance. However,
441 | nothing other than this License grants you permission to propagate or
442 | modify any covered work. These actions infringe copyright if you do
443 | not accept this License. Therefore, by modifying or propagating a
444 | covered work, you indicate your acceptance of this License to do so.
445 |
446 | 10. Automatic Licensing of Downstream Recipients.
447 |
448 | Each time you convey a covered work, the recipient automatically
449 | receives a license from the original licensors, to run, modify and
450 | propagate that work, subject to this License. You are not responsible
451 | for enforcing compliance by third parties with this License.
452 |
453 | An "entity transaction" is a transaction transferring control of an
454 | organization, or substantially all assets of one, or subdividing an
455 | organization, or merging organizations. If propagation of a covered
456 | work results from an entity transaction, each party to that
457 | transaction who receives a copy of the work also receives whatever
458 | licenses to the work the party's predecessor in interest had or could
459 | give under the previous paragraph, plus a right to possession of the
460 | Corresponding Source of the work from the predecessor in interest, if
461 | the predecessor has it or can get it with reasonable efforts.
462 |
463 | You may not impose any further restrictions on the exercise of the
464 | rights granted or affirmed under this License. For example, you may
465 | not impose a license fee, royalty, or other charge for exercise of
466 | rights granted under this License, and you may not initiate litigation
467 | (including a cross-claim or counterclaim in a lawsuit) alleging that
468 | any patent claim is infringed by making, using, selling, offering for
469 | sale, or importing the Program or any portion of it.
470 |
471 | 11. Patents.
472 |
473 | A "contributor" is a copyright holder who authorizes use under this
474 | License of the Program or a work on which the Program is based. The
475 | work thus licensed is called the contributor's "contributor version".
476 |
477 | A contributor's "essential patent claims" are all patent claims
478 | owned or controlled by the contributor, whether already acquired or
479 | hereafter acquired, that would be infringed by some manner, permitted
480 | by this License, of making, using, or selling its contributor version,
481 | but do not include claims that would be infringed only as a
482 | consequence of further modification of the contributor version. For
483 | purposes of this definition, "control" includes the right to grant
484 | patent sublicenses in a manner consistent with the requirements of
485 | this License.
486 |
487 | Each contributor grants you a non-exclusive, worldwide, royalty-free
488 | patent license under the contributor's essential patent claims, to
489 | make, use, sell, offer for sale, import and otherwise run, modify and
490 | propagate the contents of its contributor version.
491 |
492 | In the following three paragraphs, a "patent license" is any express
493 | agreement or commitment, however denominated, not to enforce a patent
494 | (such as an express permission to practice a patent or covenant not to
495 | sue for patent infringement). To "grant" such a patent license to a
496 | party means to make such an agreement or commitment not to enforce a
497 | patent against the party.
498 |
499 | If you convey a covered work, knowingly relying on a patent license,
500 | and the Corresponding Source of the work is not available for anyone
501 | to copy, free of charge and under the terms of this License, through a
502 | publicly available network server or other readily accessible means,
503 | then you must either (1) cause the Corresponding Source to be so
504 | available, or (2) arrange to deprive yourself of the benefit of the
505 | patent license for this particular work, or (3) arrange, in a manner
506 | consistent with the requirements of this License, to extend the patent
507 | license to downstream recipients. "Knowingly relying" means you have
508 | actual knowledge that, but for the patent license, your conveying the
509 | covered work in a country, or your recipient's use of the covered work
510 | in a country, would infringe one or more identifiable patents in that
511 | country that you have reason to believe are valid.
512 |
513 | If, pursuant to or in connection with a single transaction or
514 | arrangement, you convey, or propagate by procuring conveyance of, a
515 | covered work, and grant a patent license to some of the parties
516 | receiving the covered work authorizing them to use, propagate, modify
517 | or convey a specific copy of the covered work, then the patent license
518 | you grant is automatically extended to all recipients of the covered
519 | work and works based on it.
520 |
521 | A patent license is "discriminatory" if it does not include within
522 | the scope of its coverage, prohibits the exercise of, or is
523 | conditioned on the non-exercise of one or more of the rights that are
524 | specifically granted under this License. You may not convey a covered
525 | work if you are a party to an arrangement with a third party that is
526 | in the business of distributing software, under which you make payment
527 | to the third party based on the extent of your activity of conveying
528 | the work, and under which the third party grants, to any of the
529 | parties who would receive the covered work from you, a discriminatory
530 | patent license (a) in connection with copies of the covered work
531 | conveyed by you (or copies made from those copies), or (b) primarily
532 | for and in connection with specific products or compilations that
533 | contain the covered work, unless you entered into that arrangement,
534 | or that patent license was granted, prior to 28 March 2007.
535 |
536 | Nothing in this License shall be construed as excluding or limiting
537 | any implied license or other defenses to infringement that may
538 | otherwise be available to you under applicable patent law.
539 |
540 | 12. No Surrender of Others' Freedom.
541 |
542 | If conditions are imposed on you (whether by court order, agreement or
543 | otherwise) that contradict the conditions of this License, they do not
544 | excuse you from the conditions of this License. If you cannot convey a
545 | covered work so as to satisfy simultaneously your obligations under this
546 | License and any other pertinent obligations, then as a consequence you may
547 | not convey it at all. For example, if you agree to terms that obligate you
548 | to collect a royalty for further conveying from those to whom you convey
549 | the Program, the only way you could satisfy both those terms and this
550 | License would be to refrain entirely from conveying the Program.
551 |
552 | 13. Use with the GNU Affero General Public License.
553 |
554 | Notwithstanding any other provision of this License, you have
555 | permission to link or combine any covered work with a work licensed
556 | under version 3 of the GNU Affero General Public License into a single
557 | combined work, and to convey the resulting work. The terms of this
558 | License will continue to apply to the part which is the covered work,
559 | but the special requirements of the GNU Affero General Public License,
560 | section 13, concerning interaction through a network will apply to the
561 | combination as such.
562 |
563 | 14. Revised Versions of this License.
564 |
565 | The Free Software Foundation may publish revised and/or new versions of
566 | the GNU General Public License from time to time. Such new versions will
567 | be similar in spirit to the present version, but may differ in detail to
568 | address new problems or concerns.
569 |
570 | Each version is given a distinguishing version number. If the
571 | Program specifies that a certain numbered version of the GNU General
572 | Public License "or any later version" applies to it, you have the
573 | option of following the terms and conditions either of that numbered
574 | version or of any later version published by the Free Software
575 | Foundation. If the Program does not specify a version number of the
576 | GNU General Public License, you may choose any version ever published
577 | by the Free Software Foundation.
578 |
579 | If the Program specifies that a proxy can decide which future
580 | versions of the GNU General Public License can be used, that proxy's
581 | public statement of acceptance of a version permanently authorizes you
582 | to choose that version for the Program.
583 |
584 | Later license versions may give you additional or different
585 | permissions. However, no additional obligations are imposed on any
586 | author or copyright holder as a result of your choosing to follow a
587 | later version.
588 |
589 | 15. Disclaimer of Warranty.
590 |
591 | THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592 | APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596 | PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597 | IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599 |
600 | 16. Limitation of Liability.
601 |
602 | IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610 | SUCH DAMAGES.
611 |
612 | 17. Interpretation of Sections 15 and 16.
613 |
614 | If the disclaimer of warranty and limitation of liability provided
615 | above cannot be given local legal effect according to their terms,
616 | reviewing courts shall apply local law that most closely approximates
617 | an absolute waiver of all civil liability in connection with the
618 | Program, unless a warranty or assumption of liability accompanies a
619 | copy of the Program in return for a fee.
620 |
621 | END OF TERMS AND CONDITIONS
622 |
623 | How to Apply These Terms to Your New Programs
624 |
625 | If you develop a new program, and you want it to be of the greatest
626 | possible use to the public, the best way to achieve this is to make it
627 | free software which everyone can redistribute and change under these terms.
628 |
629 | To do so, attach the following notices to the program. It is safest
630 | to attach them to the start of each source file to most effectively
631 | state the exclusion of warranty; and each file should have at least
632 | the "copyright" line and a pointer to where the full notice is found.
633 |
634 |
635 | Copyright (C)
636 |
637 | This program is free software: you can redistribute it and/or modify
638 | it under the terms of the GNU General Public License as published by
639 | the Free Software Foundation, either version 3 of the License, or
640 | (at your option) any later version.
641 |
642 | This program is distributed in the hope that it will be useful,
643 | but WITHOUT ANY WARRANTY; without even the implied warranty of
644 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
645 | GNU General Public License for more details.
646 |
647 | You should have received a copy of the GNU General Public License
648 | along with this program. If not, see .
649 |
650 | Also add information on how to contact you by electronic and paper mail.
651 |
652 | If the program does terminal interaction, make it output a short
653 | notice like this when it starts in an interactive mode:
654 |
655 | Copyright (C)
656 | This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657 | This is free software, and you are welcome to redistribute it
658 | under certain conditions; type `show c' for details.
659 |
660 | The hypothetical commands `show w' and `show c' should show the appropriate
661 | parts of the General Public License. Of course, your program's commands
662 | might be different; for a GUI interface, you would use an "about box".
663 |
664 | You should also get your employer (if you work as a programmer) or school,
665 | if any, to sign a "copyright disclaimer" for the program, if necessary.
666 | For more information on this, and how to apply and follow the GNU GPL, see
667 | .
668 |
669 | The GNU General Public License does not permit incorporating your program
670 | into proprietary programs. If your program is a subroutine library, you
671 | may consider it more useful to permit linking proprietary applications with
672 | the library. If this is what you want to do, use the GNU Lesser General
673 | Public License instead of this License. But first, please read
674 | .
675 |
--------------------------------------------------------------------------------
/NetCap-output/benign/firefox/TLSHandshakes.7z:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/NetCap-output/benign/firefox/TLSHandshakes.7z
--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # TLS Mal Detect
2 | Using machine learning to detect malware in encrypted TLS traffic metadata
3 |
4 | [](https://www.gnu.org/licenses/gpl-3.0)
5 | 
6 | 
7 | 
8 | 
9 | [](https://docs.python.org/3.8/)
10 | ---
11 |
12 | The purpose of this repository is to evaluate multiple machine learning algorithms and demonstrate their ability to accurately classify malicious traffic. Three models were used in this research, a One-Class Support Vector Machine, a Support Vector Machine, and an Autoencoder Neural Network. This repository is supplied as a part of a research assignment in support of my Master's of Science in Information Security Engineering from SANS Technology Institute entitled *Malware Detection in Encrypted TLS Traffic Through Machine Learning*.
13 |
14 | Visit the accompanying Jupyter Notebook Repo to step through a basic demonstration of some of the data analysis and one of the ML models used in this research.
15 |
16 | [TLS-Mal-Detect Jupyter Repository](https://github.com/1computerguy/tls-mal-detect-jupyter)
17 |
18 | [TLS-Mal-Detect Jupyter Notebook Binder](https://notebooks.gesis.org/binder/v2/gh/1computerguy/tls-mal-detect-jupyter/HEAD?filepath=anomaly-detect.ipynb)
19 |
20 | ---
21 |
22 | > NOTE: If you decide to use this program with Windows, there are issues with pathlib. It works best if you convert the pathlib paths to raw strings.
23 |
24 | ## Run from Docker
25 |
26 | The easiest way to use this is to use Docker.
27 | 1. Download repository (git clone https://github.com/1computerguy/tls-mal-detect)
28 | 2. cd to tls-mal-detect directory
29 | 3. Unzip the file `test-train-data/test_train_data.7z`
30 | 4. Build the container with Docker:
31 | - Change directory to the tls-mal-detect repo directory
32 | - Run the docker build command
33 | - Run the docker container
34 |
35 | ```
36 | docker build . --tag tls-mal-detect:latest
37 | docker run --rm -it tls-mal-detect
38 | ```
39 |
40 | The `docker run` command above will provide the script help documentation below:
41 | ```
42 | usage: anomaly-detect.py [-h] -d DATA_SIZE [-m MALWARE_SIZE] [-t TEST_SIZE]
43 | [-o ML_MODEL] [-s] [-l] [-f FILE] [-r] [-g GRAPH]
44 | [-p] [-e] [-c CSV_FILE]
45 |
46 | Run an ML model to analyse TLS data for malicious activity.
47 |
48 | optional arguments:
49 | -h, --help show this help message and exit
50 | -d DATA_SIZE, --data DATA_SIZE
51 | Data sample size to analyze
52 | -m MALWARE_SIZE, --malware MALWARE_SIZE
53 | Percentage of dataset that is Malware
54 | -t TEST_SIZE, --test TEST_SIZE
55 | Percentage of dataset to use for validation
56 | -o ML_MODEL, --model ML_MODEL
57 | Machine Learning model to use. Acceptable values are:
58 | - ae = Autoencoder
59 | - svm = Support Vector Machine
60 | - oc-svm = One-Class SVM
61 | -s, --save Save the trained model - REQUIRES the -f/--file option
62 | -l, --load Evaluate data against a trained model - REQUIRES the -f/--file option
63 | -f FILE, --file FILE Save/Load file path
64 | -r, --scores Print 10-fold cross-validated Accuracy, Recall, Precision, and F2 scores
65 | -g GRAPH, --graph GRAPH
66 | Visualize the modeled dataset. Acceptable values are:
67 | SVM and OC-SVM graphs:
68 | - confusion
69 | - margin (SVM only)
70 | - boundary (SVM only)
71 | - auc
72 | Autoencoder graphs:
73 | - confusion
74 | - loss
75 | - mae
76 | - thresh
77 | - scatter
78 | -p, --print Print dataset
79 | -e, --export This will save the graph to a file - REQUIRED if running in a container
80 | -c CSV_FILE, --csv CSV_FILE
81 | Location of the CSV Data file
82 | ```
83 |
84 | ---
85 |
86 | #### Examples running the script from Docker
87 |
88 |
89 | Train the One-Class SVM with a 25,000 sample dataset, a 1% malware distribution, a 20% Validation dataset, and print the cross-validated accuracy, precision, recall, and F2-scores
90 |
91 | ```
92 | docker run --rm -it tls-mal-detect -d 25000 -m 1 --model oc-svm --scores
93 | ```
94 |
95 | Run the SVM using the pre-saved model
96 |
97 | ```
98 | docker run --rm -it tls-mal-detect -d 5000 -m 20 --model svm --scores --load --file /detect/models/svm.pkl
99 | ```
100 |
101 | Run the Autoencoder and export the Confusion Matrix graph as a .png file to the current working directory of the host or VM
102 |
103 | ```
104 | docker run --rm -it -v $(pwd):/detect/graph tls-mal-detect -d 25000 -m 5 --model ae --graph confusion
105 | ```
106 |
107 | ---
108 |
109 | ## Run from Host (or VM) - *Linux recommended*
110 | 1. Download repository (git clone https://github.com/1computerguy/tls-mal-detect)
111 | 2. cd to tls-mal-detect directory
112 | 3. Unzip the file `test-train-data/test_train_data.7z`
113 | 4. Install Python3 and requirements
114 | - Install python3 according to your Operating System's requirements
115 | - Install pip
116 | - Use pip to install additional requirements `pip install -r requirements.txt`
117 | - Run the program
118 |
119 | Print script help:
120 |
121 | ```
122 | python3 anomaly-detect.py -h
123 | ```
124 |
125 | Train the One-Class SVM with a 25,000 sample dataset, a 1% malware distribution, a 20% Validation dataset, and print the cross-validated accuracy, precision, recall, and F2-scores
126 |
127 | ```
128 | python3 anomaly-detect.py -d 25000 -m 1 --model oc-svm --scores
129 | ```
130 |
131 | Run the SVM using the pre-saved model
132 |
133 | ```
134 | python3 anomaly-detect.py -d 50000 -m 20 --model svm --scores --load --file ./models/svm.pkl
135 | ```
136 |
137 | Run the Autoencoder and open the Confusion Matrix graph
138 |
139 | ```
140 | python3 anomaly-detect.py -d 25000 -m 5 --model ae --graph confusion
141 | ```
142 |
143 | Run the SVM and save the margin graph to disk
144 |
145 | ```
146 | python3 anomaly-detect.py -d 25000 -m 20 --model svm --graph margin --export
147 | ```
148 |
--------------------------------------------------------------------------------
/anomaly-detect.py:
--------------------------------------------------------------------------------
1 | #!/usr/bin/env python3
2 |
3 | import pandas as pd
4 | import matplotlib.pyplot as plt
5 | import seaborn as sns
6 | import numpy as np
7 | import joblib
8 | import os
9 |
10 | from argparse import ArgumentParser, RawTextHelpFormatter
11 | from pathlib import Path
12 | from sklearn.preprocessing import MinMaxScaler
13 | from sklearn.decomposition import PCA
14 | from sklearn.model_selection import train_test_split, cross_val_score
15 | from sklearn.svm import SVC, OneClassSVM
16 | from sklearn.metrics import confusion_matrix, roc_curve, auc, roc_auc_score
17 | from sklearn.metrics import fbeta_score, accuracy_score, precision_score, recall_score
18 | from mlxtend.plotting import plot_decision_regions
19 | from tensorflow.keras.layers import Input, Dense
20 | from tensorflow.keras.models import Sequential, load_model, Model
21 | from tensorflow.keras import regularizers
22 | from tensorflow.random import set_seed
23 | from numpy.random import seed
24 |
25 | def create_graph(x_data, y_data, graph, model=None, ae=False, occ=False, label_data=None, export=False, graph_file=Path('./graph/graph.png')):
26 | '''
27 | Graph an ML model's training or analysis output to visualize its efficacy and functionality
28 | - ae and label_data arguments are used to signify Autoencoder graphs
29 | '''
30 | # Generate confusion matrix for output data
31 | if graph == 'confusion':
32 | if ae:
33 | pred_x = [1 if e > y_data else 0 for e in x_data['Loss_mae'].values]
34 | conf_matrix = confusion_matrix(label_data, pred_x)
35 | else:
36 | if occ:
37 | ben = 0
38 | mal = 1
39 | x_data[x_data == 1] = ben
40 | x_data[x_data == -1] = mal
41 | y_data = y_data.values
42 | y_data[y_data == 1] = ben
43 | y_data[y_data == -1] = mal
44 |
45 | conf_matrix = confusion_matrix(y_data, x_data)
46 |
47 | plt.figure(figsize=(8, 6))
48 | sns.heatmap(conf_matrix,
49 | xticklabels=['Benign', 'Malware'],
50 | yticklabels=['Benign', 'Malware'],
51 | annot=True, fmt='d')
52 |
53 | plt.title('Confusion Matrix')
54 | plt.ylabel('True class')
55 | plt.xlabel('Predicted class')
56 | # Create an SVM margin graph to visualize the Maximal Margin
57 | elif graph == 'margin':
58 | x_data = np.array(x_data)
59 |
60 | plt.scatter(x_data[:, 0], x_data[:, 1], c=y_data.values, s=30, cmap=plt.cm.winter, alpha=0.5)
61 | ax = plt.gca()
62 | xlim = [-8, 6]
63 | ylim = [-3, 3]
64 | xx = np.linspace(xlim[0], xlim[1], 30)
65 | yy = np.linspace(ylim[0], ylim[1], 30)
66 | XX, YY = np.meshgrid(yy, xx)
67 | xy = np.vstack([XX.ravel(), YY.ravel()]).T
68 | Z = model.decision_function(xy).reshape(XX.shape)
69 |
70 | ax.contour(XX, YY, Z, colors='k', levels=[-1, 0, 1], alpha=0.5, linestyles=['--', '-', '--'])
71 | ax.scatter(model.support_vectors_[:, 0], model.support_vectors_[:, 1],
72 | s=100, linewidth=1, facecolors='none', edgecolors='k')
73 | # Create a colored Margin graph
74 | elif graph == 'boundary':
75 | x_data = np.array(x_data) if isinstance(x_data, pd.DataFrame) else x_data
76 | y_data = np.array(y_data).astype(np.integer)
77 | plot_decision_regions(X=x_data, y=y_data, clf=model, legend=2)
78 | plt.xlabel("Component 1", size=12)
79 | plt.ylabel("Component 2", size=12)
80 | # Graph the AUC
81 | elif graph == 'auc':
82 | fpr, tpr, _ = roc_curve(y_data, x_data)
83 | auc = roc_auc_score(y_data, x_data)
84 | plt.plot([0,1], [0,1], color='navy', lw=2, linestyle='--')
85 | plt.xlim([-0.05, 1.0])
86 | plt.ylim([0.0, 1.05])
87 | plt.xlabel('False Positive Rate')
88 | plt.ylabel('True Positive Rate')
89 | plt.plot(fpr, tpr, label='ROC Curve (area = {})'.format(str(auc)), color='darkorange')
90 | plt.title('Receiver Operating Characteristic (ROC curve)')
91 | plt.legend(loc='lower right')
92 | ###################################################
93 | # Below graphs are specifically for the Autoencoder
94 | # Graph the AE loss curve
95 | elif graph == 'loss':
96 | plt.plot(x_data,
97 | 'b',
98 | label='Training loss')
99 | plt.plot(y_data,
100 | 'r',
101 | label='Validation loss')
102 | plt.legend(loc='upper right')
103 | plt.xlabel('Epochs')
104 | plt.ylabel('Loss, [mse]')
105 | plt.ylim([0,.014])
106 | # Show the Mean Absolute Error graph (used for the threshold)
107 | # to determine the number of standard deviations above the peak of the
108 | # curve that the learning reaches "0"
109 | elif graph == 'mae':
110 | X_pred = model.predict(np.array(x_data))
111 | X_pred = pd.DataFrame(X_pred, columns=x_data.columns)
112 | X_pred.index = x_data.index
113 |
114 | pred = pd.DataFrame(index=x_data.index)
115 | pred['Loss_mae'] = np.mean(np.abs(X_pred - x_data), axis = 1)
116 | plt.figure()
117 | sns.distplot(pred['Loss_mae'],
118 | bins = 10,
119 | kde= True,
120 | color = 'blue')
121 | plt.xlim([0.0,.02])
122 | # Simple time-based plot of threshold and data points
123 | elif graph == 'thresh':
124 | x_data.plot(logy=True, figsize = (10,6), ylim = [1e-2,1e0], color = ['blue','red'])
125 | # Scatter chart offering clearer, easier to understand malicious vs. benign data
126 | # representation with threshold marker
127 | elif graph == 'scatter':
128 | sns.scatterplot( x=x_data.index, y='Loss_mae', data=x_data, palette='plasma', hue='Anomaly')
129 | plt.axhline(y=y_data)
130 |
131 | if export:
132 | plt.savefig(graph_file)
133 | else:
134 | plt.show()
135 |
136 | def f_beta(beta, precision, recall):
137 | '''
138 | Calculate the F score indicating the beta value to use 0.5, 1, or 2
139 | '''
140 | return (beta*beta + 1) * precision * recall / (beta * beta * precision + recall)
141 |
142 | def save_model(model, filename, ae=False):
143 | '''
144 | Save a model to disk for later analysis
145 | '''
146 | model_output_path = os.path.split(os.path.abspath(filename))[0]
147 | if not os.path.exists(model_output_path):
148 | os.mkdir(model_output_path)
149 |
150 | if ae:
151 | model.save(filename)
152 | else:
153 | joblib.dump(model, filename)
154 |
155 | def import_model(filename, ae=False):
156 | '''
157 | Import a pre-trained model for analysis
158 | '''
159 | if os.path.exists(filename):
160 | if ae:
161 | loaded_model = load_model(filename)
162 | else:
163 | loaded_model = joblib.load(filename)
164 | else:
165 | loaded_model = 'The file path {} does not exist...'.format(filename)
166 |
167 | return loaded_model
168 |
169 | def AE_threshold(train_dist, pred_dist, extreme=False):
170 | '''
171 | Calculate the Autoencoder threshold
172 | - Above the threshold marks anomalous/malicious data
173 | - Below the threshold marks normal/benign data
174 | '''
175 | k = 4. if extreme else 3.
176 | train_thresh = np.mean(np.mean(np.abs(train_dist), axis = 1))
177 | pred_thresh = np.mean(np.mean(np.abs(pred_dist), axis = 1))
178 | threshold = np.abs(pred_thresh - train_thresh) * k
179 |
180 | return threshold
181 |
182 | def svm(data, scores=False, save=False, load=False, filename=Path('./models/svm.pkl'), graph=None, graph_file=None):
183 | '''
184 | Use a Support Vector Machine to classify malware and benign TLS traffic based on metadata
185 | gathered during the client/server handshake.
186 | '''
187 | label = 'malware_label'
188 | tt_features = data.drop(label, axis=1)
189 | tt_labels = data[label]
190 |
191 | # Feature reduction to 2 components required for margin and boundary graphs
192 | if graph == 'margin' or graph == 'boundary':
193 | pca = PCA(n_components=2).fit_transform(tt_features)
194 | tt_features = pd.DataFrame(pca)
195 |
196 | # Laod model from file
197 | if load:
198 | svclassifier = import_model(filename.absolute())
199 | feature_test = tt_features
200 | label_test = tt_labels
201 |
202 | # Train model
203 | else:
204 | feature_train, feature_test, label_train, label_test = train_test_split(tt_features, tt_labels, test_size=0.20)
205 | svclassifier = SVC(kernel='rbf', C=100, gamma=0.1, probability=True, random_state=42)
206 | svclassifier.fit(feature_train, label_train)
207 |
208 | # Save model to file
209 | if save:
210 | save_model(svclassifier, filename)
211 |
212 | # Perform n-fold cross validation and calculate the mean score across the cv=## folds
213 | if scores:
214 | print('\nCalculating SVM scores...')
215 | for val in ['accuracy', 'precision', 'recall']:
216 | score = cross_val_score(svclassifier, feature_test, label_test, cv=10, scoring=val).mean()
217 | print("{}: {}".format(val, score))
218 | if val == 'precision':
219 | prec = score
220 | elif val == 'recall':
221 | rec = score
222 | print("F2 Score: {}".format(f_beta(2.0, prec, rec)))
223 |
224 | svm_pred = svclassifier.predict(feature_test)
225 |
226 | export = True if graph_file else False
227 |
228 | if graph == 'margin' or graph == 'boundary':
229 | create_graph(tt_features, tt_labels, graph, svclassifier, export=export, graph_file=graph_file)
230 | elif graph == 'loss' or graph == 'scatter' or graph == 'mae':
231 | print('These graphs do not apply to this ML model. Try the Autoencoder to view them.')
232 | elif graph:
233 | create_graph(svm_pred, label_test, graph, export=export, graph_file=graph_file)
234 |
235 | def oc_svm(data, mal_percent, scores=False, save=False, load=False, filename=Path('./models/oc-svm.pkl'), graph=None, graph_file=None):
236 | '''
237 | Use a One-Class Support Vector Machine to classify malware and benign TLS traffic based
238 | on metadata gathered during the client/server handshake.
239 | '''
240 | # Set nu and gamma hyperparameters and test percentage
241 | test_percent = 0.20
242 | nu_value = (mal_percent / 100) * test_percent
243 | gamma_val = 0.1
244 | label = 'malware_label'
245 |
246 | #if graph == 'margin' or graph == 'boundary':
247 | # Feature reduction to 2 components required for margin and boundary graphs
248 | # OC-SVM does not graph well using any of the attempted feature reduction techniques.
249 | # I attempted PCA, Autoencoder, T-SNE, UMAP, Factor Analysis, Random Forest. The PCA
250 | # and AE are left below merely for reference.
251 |
252 | #label_data = data.malware_label
253 | #feature_data = data.drop(label, axis=1)
254 |
255 | #pca = PCA(n_components=2).fit_transform(feature_data)
256 | #data = pd.DataFrame(pca)
257 | #data = pd.concat([data, label_data], axis=1)
258 |
259 | #ae = Sequential()
260 | #ae.add(Dense(100, activation='elu',
261 | # kernel_initializer='glorot_uniform',
262 | # input_shape=(feature_data.shape[1],),
263 | # kernel_regularizer=regularizers.l2(0.0)))
264 | #ae.add(Dense(2, activation='elu', name='bottleneck', kernel_initializer='glorot_uniform'))
265 | #ae.add(Dense(100, activation='elu', kernel_initializer='glorot_uniform'))
266 | #ae.add(Dense(feature_data.shape[1], activation='sigmoid', kernel_initializer='glorot_uniform'))
267 | #ae.compile(loss='mse',optimizer='adam')
268 | #ae.fit(np.array(feature_data), np.array(feature_data), batch_size=64, epochs=20, verbose=0)
269 | #encoder = Model(ae.input, ae.get_layer('bottleneck').output)
270 | #ae_data = encoder.predict(feature_data)
271 | #data = pd.DataFrame(ae_data)
272 | #data = pd.concat([data, label_data], axis=1)
273 |
274 | # Laod model from file
275 | if load:
276 | svclassifier = import_model(filename)
277 | oc_test = data.drop(label, axis=1)
278 | oc_test_label = data[label]
279 |
280 | # Train model
281 | else:
282 | oc_benign = data[data.malware_label == 1]
283 | oc_malware = data[data.malware_label == -1]
284 |
285 | oc_b_train, oc_b_test = train_test_split(oc_benign, test_size=test_percent, random_state=1)
286 | oc_b_train = oc_b_train.drop(label, axis=1)
287 |
288 | oc_test = oc_b_test.append(oc_malware)
289 | oc_test_label = oc_test.malware_label
290 | oc_test = oc_test.drop(label, axis=1)
291 |
292 | svclassifier = OneClassSVM(nu=nu_value, kernel='rbf', gamma=gamma_val)
293 | svclassifier.fit(oc_b_train)
294 |
295 | # Save model to file
296 | if save:
297 | save_model(svclassifier, filename)
298 |
299 | # Perform n-fold cross validation and calculate the mean score across the cv=## folds
300 | if scores:
301 | print('\nCalculating OC-SVM scores...')
302 | for val in ['accuracy', 'precision', 'recall']:
303 | score = cross_val_score(svclassifier, oc_test, oc_test_label, cv=2, scoring=val).mean()
304 | print("{}: {}".format(val, score))
305 | if val == 'precision':
306 | prec = score
307 | elif val == 'recall':
308 | rec = score
309 | print("F2 Score: {}".format(f_beta(2.0, prec, rec)))
310 |
311 | oc_pred = svclassifier.predict(oc_test)
312 |
313 | export = True if graph_file else False
314 |
315 | # You can uncomment the below if statement (and change the second if graph to an elif graph)
316 | # ONLY if you enable one of the feature reduction techniques above - either PCA or the AE
317 | #if graph == 'margin' or graph == 'boundary':
318 | # create_graph(oc_test, oc_test_label, graph, svclassifier, export, graph_file)
319 | if graph == 'confusion' or graph == 'auc':
320 | create_graph(oc_pred, oc_test_label, graph, svclassifier, export=export, graph_file=graph_file, occ=True)
321 | elif graph == 'margin' or graph == 'boundary':
322 | print('You need to uncomment one of the feature reduction techniques in this function to use that graph type...')
323 | elif graph == 'loss' or graph == 'scatter' or graph == 'mae':
324 | print('These graphs do not apply to this ML model. Try the Autoencoder to view them.')
325 |
326 | def ae(data, scores=False, save=False, load=False, filename=Path('./models/ae.h5'), graph=None, graph_file=None):
327 | '''
328 | Use an Autoencoder Neural Network to classify malware and benign TLS traffic based
329 | on metadata gathered during the client/server handshake.
330 | '''
331 | seed(1)
332 | set_seed(2)
333 | NUM_EPOCHS=200
334 | BATCH_SIZE=32
335 | label = 'malware_label'
336 | act_func = 'elu'
337 | label_data = data.malware_label
338 |
339 | # Laod model from file
340 | if load:
341 | data = data.drop(label, axis=1)
342 |
343 | model = import_model(filename, True)
344 | predictions = model.predict(np.array(data))
345 | predictions = pd.DataFrame(predictions, columns=data.columns)
346 | predictions.index = data.index
347 |
348 | # Calculate threshold and provide anomaly output predictions as a dataframe
349 | threshold = AE_threshold(data, predictions)
350 | scored = pd.DataFrame(index=data.index)
351 | scored['Loss_mae'] = np.mean(np.abs(predictions-data), axis=1)
352 | scored['Threshold'] = threshold
353 | scored['Anomaly'] = scored['Loss_mae'] > scored['Threshold']
354 |
355 | # Train model
356 | else:
357 | x_train = data[data.malware_label == 0]
358 | x_train = x_train.drop(label, axis=1)
359 | x_test = data[data.malware_label == 1]
360 | x_test = x_test.drop(label, axis=1)
361 |
362 | begin_end_length = x_train.shape[1]
363 | stage_two_length = 300
364 | stage_three_length = 100
365 | stage_four_length = 2
366 |
367 | # Build AE network
368 | model = Sequential()
369 |
370 | model.add(Dense(stage_two_length, activation=act_func,
371 | kernel_initializer='glorot_uniform',
372 | kernel_regularizer=regularizers.l2(0.0),
373 | input_shape=(begin_end_length,)
374 | )
375 | )
376 |
377 | model.add(Dense(stage_three_length , activation=act_func,
378 | kernel_initializer='glorot_uniform'))
379 | model.add(Dense(stage_four_length , activation=act_func,
380 | kernel_initializer='glorot_uniform'))
381 | model.add(Dense(stage_three_length , activation=act_func,
382 | kernel_initializer='glorot_uniform'))
383 |
384 | model.add(Dense(stage_two_length , activation=act_func,
385 | kernel_initializer='glorot_uniform'))
386 |
387 | model.add(Dense(begin_end_length,
388 | kernel_initializer='glorot_uniform'))
389 |
390 | model.compile(loss='mse',optimizer='adam')
391 |
392 | history = model.fit(np.array(x_train), np.array(x_train),
393 | batch_size=BATCH_SIZE,
394 | epochs=NUM_EPOCHS,
395 | validation_split=0.05,
396 | verbose = 1)
397 |
398 | # Save model to file
399 | if save:
400 | save_model(model, filename, True)
401 |
402 | predictions = model.predict(np.array(x_test))
403 | predictions = pd.DataFrame(predictions,
404 | columns=x_test.columns)
405 | predictions.index = x_test.index
406 |
407 | # Calculate scores if conducting training and validation
408 | # - First set of measurements are for benign traffic
409 | # - Second set of measurements are for predictions based on first set
410 | # and for detecting malware
411 | scored = pd.DataFrame(index=x_test.index)
412 | threshold = AE_threshold(x_train, predictions, True)
413 | scored['Loss_mae'] = np.mean(np.abs(predictions-x_test), axis = 1)
414 | scored['Threshold'] = threshold
415 | scored['Anomaly'] = scored['Loss_mae'] > scored['Threshold']
416 |
417 | predictions_train = model.predict(np.array(x_train))
418 | predictions_train = pd.DataFrame(predictions_train,
419 | columns=x_train.columns)
420 | predictions_train.index = x_train.index
421 |
422 | scored_train = pd.DataFrame(index=x_train.index)
423 | scored_train['Loss_mae'] = np.mean(np.abs(predictions_train-x_train), axis = 1)
424 | scored_train['Threshold'] = threshold
425 | scored_train['Anomaly'] = scored_train['Loss_mae'] > scored_train['Threshold']
426 | scored = pd.concat([scored_train, scored]).sort_index()
427 |
428 | # Cross_val_score does not support AE model, so we calculate scores individually
429 | # Did not find a suitable method of N-fold cross validation...
430 | if scores:
431 | print('\nPrinting Autoencoder scores...')
432 | print('Accuracy: {}'.format(accuracy_score(label_data, scored['Anomaly'])))
433 | print('Precision: {}'.format(precision_score(label_data, scored['Anomaly'])))
434 | print('Recall: {}'.format(recall_score(label_data, scored['Anomaly'])))
435 | print('F2 Score: {}'.format(fbeta_score(label_data, scored['Anomaly'], beta=2.0)))
436 |
437 | export = True if graph_file else False
438 |
439 | if graph == 'loss':
440 | create_graph(history.history['loss'], history.history['val_loss'], graph, export=export, graph_file=graph_file)
441 | elif graph == 'scatter':
442 | create_graph(scored, threshold, graph, model, export=export, graph_file=graph_file)
443 | elif graph == 'confusion':
444 | create_graph(scored, threshold, graph, export=export, graph_file=graph_file, ae=True, label_data=label_data)
445 | elif graph == 'mae':
446 | create_graph(x_train, label_data, graph, model, export=export, graph_file=graph_file)
447 | elif graph:
448 | create_graph(scored, label_data, graph, model, export=export, graph_file=graph_file)
449 |
450 | def get_data(csv_data_file, sample_size, mal_percent=20, test_percent=20, occ=False):
451 | rand_state_val = 42
452 | full_dataset = pd.read_csv(csv_data_file)
453 |
454 | # Scale data to 0-1 value for more efficient ML analysis
455 | mm_data = MinMaxScaler().fit_transform(full_dataset)
456 | full_dataset = pd.DataFrame(mm_data, columns=full_dataset.columns)
457 |
458 | # If model is OC-SVM convert label values to 1 and -1 (this is how OC-SVM
459 | # outputs predictions, so validation requires these values
460 | if occ:
461 | label = 'malware_label'
462 | ben = 1
463 | mal = -1
464 | full_dataset.loc[full_dataset[label] == 1, label] = mal
465 | full_dataset.loc[full_dataset[label] == 0, label] = ben
466 |
467 | benign = full_dataset[full_dataset.malware_label == ben]
468 | malware = full_dataset[full_dataset.malware_label == mal]
469 |
470 | test_size = int((test_percent / 100) * sample_size)
471 | mal_size = int((mal_percent / 100) * test_size)
472 |
473 | # Prevent malware sample size from being larger than actual sample size
474 | if mal_size > malware.shape[0]:
475 | mal_size = malware.shape[0]
476 |
477 | malware = malware.sample(n=mal_size, random_state=rand_state_val)
478 |
479 | # Prevent total sample size from being larger than actual sample size
480 | total_sample_size = sample_size - mal_size
481 | if total_sample_size > benign.shape[0]:
482 | total_sample_size = benign.shape[0]
483 |
484 | benign = benign.sample(n=total_sample_size, random_state=rand_state_val)
485 | sampled_data = benign.append(malware).reset_index(drop=True)
486 | # If not OC-SVM then provide desitnated testing and malware distributions
487 | else:
488 | ben = 0
489 | mal = 1
490 |
491 | benign = full_dataset[full_dataset.malware_label == ben]
492 | malware = full_dataset[full_dataset.malware_label == mal]
493 |
494 | mal_size = int((mal_percent / 100) * sample_size)
495 |
496 | # Prevent malware sample size from being larger than actual sample size
497 | if mal_size > malware.shape[0]:
498 | mal_size = malware.shape[0]
499 |
500 | malware = malware.sample(n=mal_size, random_state=rand_state_val)
501 |
502 | # Prevent total sample size from being larger than actual sample size
503 | total_sample_size = sample_size - mal_size
504 | if total_sample_size > benign.shape[0]:
505 | total_sample_size = benign.shape[0]
506 |
507 | benign = benign.sample(n=total_sample_size, random_state=rand_state_val)
508 | sampled_data = benign.append(malware).reset_index(drop=True)
509 |
510 | sampled_data = sampled_data.sample(frac=1).reset_index(drop=True)
511 |
512 | return sampled_data
513 |
514 | def main():
515 | '''
516 | Execute above functions and run through the various ML models outlined in the paper:
517 | Malware Detection in Encrypted TLS Traffic Through Machine Learning
518 | '''
519 |
520 | parser = ArgumentParser(description='Run an ML model to analyse TLS data for malicious activity.',
521 | formatter_class=RawTextHelpFormatter)
522 | parser.add_argument('-d', '--data', action='store', dest='data_size', default=0,
523 | help='Data sample size to analyze', type=int, required=True)
524 | parser.add_argument('-m', '--malware', action='store', dest='malware_size', default=20,
525 | help='Percentage of dataset that is Malware', type=float, required=False)
526 | parser.add_argument('-t', '--test', action='store', dest='test_size', default=20,
527 | help='Percentage of dataset to use for validation', type=float, required=False)
528 | parser.add_argument('-o', '--model', action='store', dest='ml_model',
529 | help='''Machine Learning model to use. Acceptable values are:
530 | - ae = Autoencoder
531 | - svm = Support Vector Machine
532 | - oc-svm = One-Class SVM''',
533 | required=False)
534 | parser.add_argument('-s', '--save', action='store_true', dest='save_model', default=False,
535 | help='Save the trained model - REQUIRES the -f/--file option', required=False)
536 | parser.add_argument('-l', '--load', action='store_true', dest='load_model', default=False,
537 | help='Evaluate data against a trained model - REQUIRES the -f/--file option', required=False)
538 | parser.add_argument('-f', '--file', action='store', dest='file', default=None,
539 | help='Save/Load file path', required=False)
540 | parser.add_argument('-r', '--scores', action='store_true', dest='scores', default=False,
541 | help='Print 10-fold cross-validated Accuracy, Recall, Precision, and F2 scores', required=False)
542 | parser.add_argument('-g', '--graph', action='store', dest='graph', default=None,
543 | help='''Visualize the modeled dataset. Acceptable values are:
544 | SVM and OC-SVM graphs:
545 | - confusion
546 | - margin (SVM only)
547 | - boundary (SVM only)
548 | - auc
549 | Autoencoder graphs:
550 | - confusion
551 | - loss
552 | - mae
553 | - thresh
554 | - scatter''',
555 | required=False)
556 | parser.add_argument('-p', '--print', action='store_true', dest='print_data', default=False,
557 | help='Print dataset', required=False)
558 | parser.add_argument('-e', '--export', action='store_true', dest='export', default=False,
559 | help='This will save the graph to a file - REQUIRED if running in a container', required=False)
560 | parser.add_argument('-c', '--csv', action='store', dest='csv_file',
561 | help='Location of the CSV Data file', required=False)
562 |
563 | options = parser.parse_args()
564 |
565 | if (options.save_model or options.load_model) and not options.file:
566 | print('If you want to save or load a model, you must also use the -f or\n--file option and provide the location of the file.')
567 | quit()
568 |
569 | data_size = options.data_size
570 | malware_size = options.malware_size
571 | test_size = options.test_size
572 | save = options.save_model
573 | load = options.load_model
574 | scores = options.scores
575 | graph = options.graph
576 | print_data = options.print_data
577 | model = options.ml_model
578 | export_graph = options.export
579 | csv_file = options.csv_file
580 |
581 | occ = True if model == 'oc-svm' else False
582 | filename = Path(options.file) if options.file else None
583 |
584 | docker = bool(os.environ.get('RUNNING_IN_DOCKER', False))
585 |
586 | if docker:
587 | graph_file = Path('/detect/graph/{}-{}.png'.format(model, graph)) if export_graph else None
588 | csv_file = Path('/detect/data/test_train_data.csv') if not csv_file else csv_file
589 | else:
590 | graph_file = Path('./graph/{}-{}.png'.format(model, graph)) if export_graph else None
591 | csv_file = Path('./test-train-data/test_train_data.csv') if not csv_file else csv_file
592 |
593 | if load and not filename.exists():
594 | print('\n The file {} cannot be found... Please check your spelling and try again'.format(filename))
595 | quit()
596 |
597 | dataset = get_data(csv_file, data_size, malware_size, test_size, occ)
598 |
599 | if print_data:
600 | print(dataset)
601 |
602 | if model == 'ae':
603 | ae(dataset, scores, save, load, filename, graph, graph_file)
604 | elif model == 'svm':
605 | svm(dataset, scores, save, load, filename, graph, graph_file)
606 | elif model == 'oc-svm':
607 | oc_svm(dataset, malware_size, scores, save, load, filename, graph, graph_file)
608 | elif model:
609 | print('\nPlease choose a model of type ae, svm, or oc-svm... To get help using this script use the -h or --help option')
610 | quit()
611 |
612 | if __name__ == '__main__':
613 | main()
614 |
615 |
--------------------------------------------------------------------------------
/feature-reduction-tests/features.py:
--------------------------------------------------------------------------------
1 | import pandas as pd
2 | import seaborn as sns
3 | import matplotlib.pyplot as plt
4 | import numpy as np
5 | import tensorflow as tf
6 | import pingouin as pg
7 | import matplotlib.patches as mpatches
8 |
9 | from scipy.stats import bartlett, levene
10 | from tensorflow.keras.layers import Input, Dense
11 | from tensorflow.keras.models import Model
12 | from tensorflow.keras import regularizers
13 | from tensorflow.random import set_seed
14 | from statsmodels.stats.outliers_influence import variance_inflation_factor
15 | from sklearn.preprocessing import MinMaxScaler, StandardScaler
16 | from sklearn.decomposition import PCA
17 | from sklearn.model_selection import train_test_split
18 | from tensorflow.keras.callbacks import ModelCheckpoint, TensorBoard
19 | from sklearn.metrics import confusion_matrix, precision_recall_curve, auc, roc_curve
20 | from numpy.random import seed
21 | from sklearn.ensemble import RandomForestClassifier
22 |
23 |
24 | def malware_distribution(data, label):
25 | '''
26 | Generate a bar chart showing malware to benign data ratio
27 | '''
28 | # Specify data column to calculate
29 | target = data[label]
30 | # Get dataset length for percentage calculation
31 | total = len(data)
32 | # Define graph area and title
33 | plt.figure(figsize = (6, 6))
34 | plt.title("Malware Dataset Distribution")
35 |
36 | # Generate count plot and turn into bar graph for display
37 | ax = sns.countplot(target)
38 | for p in ax.patches:
39 | percentage = '{:.0f}%'.format(p.get_height() / total * 100)
40 | x = p.get_x() + p.get_width() / 2
41 | y = p.get_height() + 5
42 | ax.annotate(percentage, (x, y), ha = 'center')
43 |
44 | plt.show()
45 |
46 | def dataset_heatmap (data, label, annotate=False):
47 | '''
48 | Analyze all features together in a large heatmap. This is nearly impossible to interpret using all 519 features.
49 | '''
50 | #Gausidan distrabution of dataset
51 | data_features = data.drop(label, axis=1)
52 | data_standard_dev = (data_features - data_features.mean()) / data_features.std()
53 | gaussian_data = pd.concat([data[label], data_standard_dev], axis=1)
54 |
55 | # Define plot area
56 | plt.figure(figsize = (10, 8))
57 | plt.title("Correlation Heatmap")
58 | correlation = gaussian_data.corr()
59 | if annotate:
60 | sns.heatmap(correlation, annot = annotate, fmt = '.2f', cmap = 'coolwarm')
61 | else:
62 | sns.heatmap(correlation, annot = annotate, cmap = 'coolwarm')
63 |
64 | plt.show()
65 |
66 | def calculate_vif (data, label):
67 | '''
68 | Method to calculate Variance Inflation Factor (VIF) to determine multi-collinearity of data. Had difficulty
69 | interpreting this output...
70 | '''
71 | data = data.drop(label, axis=1)
72 | vif_data = pd.DataFrame()
73 | vif_data['feature'] = data.columns
74 | vif_data['VIF'] = [variance_inflation_factor(data.values, i) for i in range(len(data.columns))]
75 |
76 | for feature in vif_data:
77 | print(feature)
78 |
79 | def mal_ben_hist (data, label, graph_set, benign_percent):
80 | '''
81 | Generate a feature for feature histogram comparing the importance of various features in distinguishing between
82 | malware and benign traffic.
83 | '''
84 | malware_label = data.malware_label
85 | # Remove malware label
86 | data = data.drop(label, axis=1)
87 | #std_data = StandardScaler().fit_transform(data)
88 | #norm_data = normalize(data)
89 | mm_data = MinMaxScaler().fit_transform(data)
90 | data = pd.DataFrame(mm_data, columns = data.columns)
91 | _, axes = plt.subplots(10, 3, figsize=(12, 9)) # 3 columns containing 10 figures
92 |
93 | begin = graph_set * 30
94 | end = begin + 30
95 | data_to_graph = data.iloc[:, begin:end]
96 |
97 | data_to_graph = pd.concat([data_to_graph, malware_label], axis=1)
98 | malware = data_to_graph[data_to_graph.malware_label == 1]
99 | benign = data_to_graph[data_to_graph.malware_label == 0]
100 | # Get a percentage of the benign data set to balance measurements for analysis
101 | # This can be changed to view graphs differently, but is very helpful to truly
102 | # see the differences between benign and malicious traffic side by side
103 | percent = int(len(benign) * (benign_percent / 100))
104 | benign = benign.sample(n=percent)
105 | ax = axes.ravel()
106 | for i in range(data_to_graph.shape[1] - 1):
107 | _, bins = np.histogram(data_to_graph.iloc[:, i], bins=40)
108 | ax[i].hist(malware.iloc[:, i], bins=bins, color='r', alpha=.5) # Red for malware
109 | ax[i].hist(benign.iloc[:, i], bins=bins, color='g', alpha=0.3) # Green for benign
110 | ax[i].set_title(data_to_graph.columns[i], fontsize=9)
111 | ax[i].axes.get_xaxis().set_visible(False) # Just want to see separation not measurements
112 | ax[i].set_yticks(())
113 |
114 | ax[0].legend(['malware', 'benign'], loc='best', fontsize=8)
115 | plt.tight_layout()
116 | plt.show()
117 |
118 |
119 | def calculate_pca (data, label, components=10, fit=False, graph=None):
120 | '''
121 | Used to generate a desired number of "Principle Components" from the input data and return the calculated
122 | output components as a pandas dataframe
123 | '''
124 | features = data.columns
125 | # Remove features from data
126 | data_vals = data.loc[:, features].values
127 | # Define the target/label
128 | data_label = data.loc[:, [label]].values
129 | # Normalize the dataset
130 | std_data = MinMaxScaler().fit_transform(data_vals)
131 |
132 | # Calculate the 10 most important components
133 | #pca = PCA(n_components = components)
134 | if fit:
135 | data_pca_vals = PCA().fit(std_data.data)
136 | else:
137 | data_pca_vals = PCA().fit_transform(std_data)
138 | #pca_dataframe = pd.DataFrame(data = data_pca_vals)
139 | #final_pca_dataframe = pd.concat([pca_dataframe, data[[label]]], axis=1)
140 |
141 | if graph == 'heatmap':
142 | correlation = final_pca_dataframe.corr()
143 | sns.heatmap(correlation, annot=True, fmt='.2f', cmap = 'coolwarm')
144 | plt.show()
145 | elif graph == 'pairplot':
146 | sns.pairplot(final_pca_dataframe, kind='scatter', hue=label, markers=['o', 's'], palette='Set2')
147 | plt.show()
148 | elif graph == 'scatter':
149 | colors = {'0': 'darkblue', '1': 'darkorange'}
150 | plt.scatter(data_pca_vals[:, 0], data_pca_vals[:, 1],
151 | c=pd.Series(data['malware_label']).astype(str).map(colors), edgecolor='none',
152 | alpha=0.5, cmap='viridis')
153 | m = mpatches.Patch(color='darkblue', label='Benign')
154 | b = mpatches.Patch(color='darkorange', label='Malware')
155 | plt.legend(handles=[m, b])
156 | plt.show()
157 | elif graph == 'comp_curve':
158 | plt.plot(np.cumsum(data_pca_vals.explained_variance_ratio_))
159 | plt.xlabel('number of componsents')
160 | plt.ylabel('cumulative explained variance')
161 | plt.show()
162 |
163 | #return pd.DataFrame(final_pca_dataframe)
164 |
165 | def random_forest(data, label, estimators, graph=None):
166 | '''
167 | Use a Random Forest to determine feature importance. Can also be used to return the top number of
168 | features determined by the "IF" statement of val_tuple[1] value. This is the threshold determined by
169 | the Random Forest regressor and can be determiend by generating and analyzing the bar chart.
170 |
171 | This proved to be between 65%-80% successful in classifing malicious traffic by itself, however, the SVM
172 | used in tls-mal-detect.py was more successful due to its robust ability to deal with outliers.
173 | '''
174 | seed(1)
175 | set_seed(2)
176 | SEED = 123
177 | DATA_SPLIT_PCT = 0.3
178 | data_label = data.malware_label
179 | data = data.drop(label, axis=1)
180 | features = data.columns
181 |
182 | # Standardize the dataset
183 | std_data = MinMaxScaler().fit_transform(data)
184 |
185 | x_train, x_test, train_labels, test_labels = train_test_split(std_data, data_label, random_state=SEED, test_size=DATA_SPLIT_PCT)
186 |
187 | regressor = RandomForestClassifier(n_estimators=estimators, random_state=SEED)
188 | regressor.fit(x_train, train_labels)
189 | predictions = regressor.predict(x_test)
190 | feat_series = regressor.feature_importances_
191 |
192 | if graph == 'bar':
193 | pd.Series(feat_series, index=features).nlargest(50).plot(kind='barh').invert_yaxis()
194 | plt.show()
195 |
196 | feature_list = [(feature, round(importance, 2)) for feature, importance in zip(list(features), list(feat_series))]
197 | top_10_feature_list = []
198 | for val_tuple in feature_list:
199 | if val_tuple[1] >= 0.03:
200 | top_10_feature_list.append(val_tuple[0])
201 |
202 | final_data = data[top_10_feature_list]
203 | final_data = pd.DataFrame(MinMaxScaler().fit_transform(final_data), columns=top_10_feature_list)
204 | final_data = pd.concat([data_label, final_data], axis=1)
205 | return final_data
206 |
207 | def autoencoded_features (data, label, final_features, graph=None):
208 | '''
209 | Attempt to analyze and interpret data using a Sparse, Stacked Autoencoder. Was not too successful,
210 | but leaving here for potential, future analysis in feature reduction in lieu of Random Forest or
211 | PCA. Read some interesting research where that was successful.
212 | '''
213 | # Balance dataset based on percentage passed to function
214 | output_vals = []
215 | seed(1)
216 | set_seed(2)
217 | SEED = 123
218 | DATA_SPLIT_PCT = 0.2
219 |
220 | # Split into training and testing datasets
221 | x_train, x_test = train_test_split(data, test_size=DATA_SPLIT_PCT, random_state=SEED)
222 | x_train, x_valid = train_test_split(x_train, test_size=DATA_SPLIT_PCT, random_state=SEED)
223 |
224 | x_train_0 = x_train.loc[data[label] == 0]
225 | x_train_1 = x_train.loc[data[label] == 1]
226 | x_train_0_x = x_train_0.drop([label], axis=1)
227 | x_train_1_x = x_train_1.drop([label], axis=1)
228 |
229 | x_valid_0 = x_valid.loc[data[label] == 0]
230 | x_valid_1 = x_valid.loc[data[label] == 1]
231 | x_valid_0_x = x_valid_0.drop([label], axis=1)
232 | x_valid_1_x = x_valid_1.drop([label], axis=1)
233 |
234 | x_test_0 = x_test.loc[data[label] == 0]
235 | x_test_1 = x_test.loc[data[label] == 1]
236 | x_test_0_x = x_test_0.drop([label], axis=1)
237 | x_test_1_x = x_test_1.drop([label], axis=1)
238 |
239 | scaler = StandardScaler().fit(x_train_0_x)
240 | x_train_0_x_rescaled = scaler.transform(x_train_0_x)
241 | x_valid_0_x_rescaled = scaler.transform(x_valid_0_x)
242 | x_valid_x_rescaled = scaler.transform(x_valid.drop([label], axis=1))
243 |
244 | x_test_0_x_rescaled = scaler.transform(x_test_0_x)
245 | x_test_x_rescaled = scaler.transform(x_test.drop([label], axis=1))
246 |
247 | # Autoencoder values
248 | learning_epochs = 200
249 | batch_size = 128
250 | input_dim = x_train_0_x_rescaled.shape[1]
251 | #input_dim = x_train_1.shape[1]
252 | encoding_dim = int(input_dim / 2)
253 | hidden_dim_1 = int(encoding_dim / 2)
254 | hidden_dim_2 = int(hidden_dim_1 / 2)
255 | final_hidden_dim = final_features
256 | learning_rate = 1e-6
257 |
258 | input_layer = Input(shape=(input_dim, ))
259 | encoder = Dense(encoding_dim, activation='relu', activity_regularizer=regularizers.l1(learning_rate))(input_layer)
260 | encoder = Dense(hidden_dim_1, activation='relu')(encoder)
261 | encoder = Dense(hidden_dim_2, activation='relu')(encoder)
262 | encoder = Dense(final_hidden_dim, activation='relu')(encoder)
263 | decoder = Dense(final_hidden_dim, activation='relu')(encoder)
264 | decoder = Dense(hidden_dim_2, activation='relu')(decoder)
265 | decoder = Dense(hidden_dim_1, activation='relu')(decoder)
266 | decoder = Dense(encoding_dim, activation='relu')(decoder)
267 | decoder = Dense(input_dim, activation='linear')(decoder)
268 | autoencoder = Model(inputs=input_layer, outputs=decoder)
269 | autoencoder.summary()
270 |
271 | autoencoder.compile(metrics=['accuracy'], loss='mean_squared_error', optimizer='adam')
272 | write_model = ModelCheckpoint(filepath=r'C:\Users\bryan\Desktop\ae_data\model\ae_calssifier.h5', save_best_only=True, verbose=0)
273 | write_logs = TensorBoard(log_dir=r'C:\Users\bryan\Desktop\ae_data\logs', histogram_freq=0, write_graph=True, write_images=True)
274 | history = autoencoder.fit(x_train_0_x_rescaled, x_train_0_x_rescaled,
275 | epochs=learning_epochs,
276 | batch_size=batch_size,
277 | validation_data=(x_valid_0_x_rescaled, x_valid_0_x_rescaled),
278 | verbose=1,
279 | callbacks=[write_model, write_logs]).history
280 |
281 | valid_x_predictions = autoencoder.predict(x_valid_x_rescaled)
282 | mse = np.mean(np.power(x_valid_x_rescaled - valid_x_predictions, 2), axis=1)
283 |
284 | error_df = pd.DataFrame({'Reconstruction_error': mse, 'True_class': x_valid[label]})
285 | false_pos_rate, true_pos_rate, thresholds = roc_curve(error_df['True_class'], error_df['Reconstruction_error'])
286 | threshold = np.mean(thresholds)
287 | threshold_fixed = float("{:0.4f}".format(threshold))
288 | roc_auc = auc(false_pos_rate, true_pos_rate,)
289 |
290 | output_vals.append('MSE: {}'.format(mse))
291 | output_vals.append('Threshold mean: {}'.format(threshold))
292 | output_vals.append('AUC: {}'.foramt(auc(false_pos_rate, true_pos_rate)))
293 |
294 | if graph == 'loss':
295 | plt.plot(history['loss'])
296 | plt.plot(history['val_loss'])
297 | plt.title('model_loss')
298 | plt.ylabel('loss')
299 | plt.xlabel('epoch')
300 | plt.legend(['train', 'test'], loc='upper left')
301 | plt.show()
302 | elif graph == 'pre_call':
303 | precision_rt, recall_rt, threshold_rt = precision_recall_curve(error_df.True_class, error_df.Reconstruction_error)
304 | plt.plot(threshold_rt, precision_rt[1:], label='Precision', linewidth=5)
305 | plt.plot(threshold_rt, recall_rt[1:], label='Recall', linewidth=5)
306 | plt.title('Precision and recall for different threshold values')
307 | plt.xlabel('Threshold')
308 | plt.ylabel('Precision/Recall')
309 | plt.legend()
310 | plt.show()
311 | elif graph == 're_error':
312 | groups = error_df.groupby('True_class')
313 | fig, ax = plt.subplots()
314 | for name, group in groups:
315 | ax.plot(group.index, group.Reconstruction_error, marker='o', ms=3.5, linestyle='', label='Malware Estimation' if name == 1 else 'Benign Estimate')
316 | ax.hlines(threshold_fixed, ax.get_xlim()[0], ax.get_xlim()[1], colors='r', zorder=100, label='Threshold')
317 | ax.legend()
318 | plt.title('Reconstruction error for malicious/benign traffic')
319 | plt.ylabel('Reconstruction error')
320 | plt.xlabel('Data point index')
321 | plt.show()
322 | elif graph == 'heatmap':
323 | pred_y = [1 if e > threshold_fixed else 0 for e in error_df['Reconstruction_error'].values]
324 | conf_matrix = confusion_matrix(error_df['True_class'], pred_y)
325 | plt.figure(figsize=(8, 6))
326 | sns.heatmap(conf_matrix,
327 | xticklabels=['Benign', 'Malware'],
328 | yticklabels=['Benign', 'Malware'],
329 | annot=True, fmt='d')
330 | plt.title('Confusion Matrix')
331 | plt.ylabel('True class')
332 | plt.xlabel('Predicted class')
333 | plt.show()
334 | elif graph == 'roc':
335 | plt.plot(false_pos_rate, true_pos_rate, linewidth=5, label='AUC = %0.3f'% roc_auc)
336 | plt.plot([0,1],[0,1], linewidth=5)
337 | plt.xlim([-0.01, 1])
338 | plt.ylim([0, 1.01])
339 | plt.legend(loc='lower right')
340 | plt.title('Reciever operating charactistic curve (ROC)')
341 | plt.ylabel('True Positive Rating')
342 | plt.xlabel('False Positive Rate')
343 | plt.show()
344 |
345 | return output_vals
346 |
347 | def factor_analysis(data, label, eq_var=True):
348 | '''
349 | Method to perform factor analysis to determine the most important features and calculate
350 | the ability to distinguish between malware and benign traffic using varying numbers of "important"
351 | features.
352 | '''
353 | np.seterr(divide='raise')
354 | for feature in data.columns:
355 | try:
356 | bart_vals = pg.homoscedasticity(data, dv=feature, group='malware_label', method='bartlett')
357 | lev_vals = pg.homoscedasticity(data, dv=feature, group='malware_label')
358 | if bart_vals['equal_var'].values[0] == eq_var:
359 | print("Bartlett Value feature: {}\nT Value: {}\nP Value: {}".format(feature, bart_vals['T'].values[0], bart_vals['pval'].values[0]))
360 | elif lev_vals['equal_var'].values[0] == eq_var:
361 | print("Levene Value feature: {}\nT Value: {}\nP Value: {}".format(feature, bart_vals['T'].values[0], bart_vals['pval'].values[0]))
362 | except Exception as e:
363 | pass
364 |
365 | #if __name__ == '__main__':
366 | #def load_input (csv_data_file):
367 | csv_data_file = r'test-train-data\test\test_train_data-all.csv'
368 | csv_data_file_new = r'test-train-data\validate\validate_data_new-all.csv'
369 | dataset = pd.read_csv(csv_data_file)
370 | new_dataset = pd.read_csv(csv_data_file_new)
371 |
372 | full_data = pd.concat([dataset, new_dataset], axis=0)
373 | full_data = full_data.reset_index()
374 | # Remove columns filled with all 0 value (these will be statistically insignifant and will cause
375 | # issues when using correlation methods of analysis)
376 | data_no_z_cols = dataset.loc[:, (dataset != 0).any(axis=0)]
377 |
378 | # NOTE TO SELF
379 | # To import as module for testing:
380 | # from importlib import import_module, reload
381 | # a = import_module('features')
382 | # Use this when you make changes
383 | # reload(a)
384 | #
385 | # Call methods using below syntax:
386 | # a.random_forest(a.dataset, 'labelname', 100, 'bar')
--------------------------------------------------------------------------------
/feature-reduction-tests/graph_data.py:
--------------------------------------------------------------------------------
1 | import pandas as pd
2 | import numpy as np
3 | import seaborn as sns
4 | import matplotlib.pyplot as plt
5 | from sklearn.preprocessing import MinMaxScaler
6 |
7 | #csv_data_file = r'test-train-data\test\test_train_data-all.csv'
8 | #new_csv_data_file = r'test-train-data\validate\validate_data_new-all.csv'
9 | csv_data_file = r'test-train-data\test_train_data.csv'
10 |
11 | #all_data = pd.concat([pd.read_csv(csv_data_file), pd.read_csv(new_csv_data_file)], axis=0)
12 | all_data = pd.read_csv(csv_data_file)
13 | all_data_nz = all_data.loc[:, (all_data != 0).any(axis=0)]
14 |
15 | mal = all_data_nz[all_data_nz.malware_label == 1]
16 | #mal = pd.DataFrame(pd.DataFrame(MinMaxScaler().fit_transform(mal), columns=mal.columns).mean(), columns=['Malware'])
17 | ben = all_data_nz[all_data_nz.malware_label == 0]
18 | #ben = pd.DataFrame(pd.DataFrame(MinMaxScaler().fit_transform(ben), columns=ben.columns).mean(), columns=['Benign'])
19 |
20 | #######################
21 | # CS Number of CS's used
22 | #sns.barplot(data=cs_sum, x=cs_sum.index)
23 | #mal_cs_nz = len(mal_cs_data.loc[:, (mal_cs_data != 0).any(axis=0)].columns)
24 | #ben_cs_nz = len(ben_cs_data.loc[:, (ben_cs_data != 0).any(axis=0)].columns)
25 |
26 | # Add labels above bars
27 | #cs_count = pd.DataFrame({'Label': ['Benign Signature Algorithms', 'Malware Signature Algorithms'], 'Count': [ben_cs_nz, mal_cs_nz]})
28 | #ax = sns.barplot(data=cs_count, y='Count', x='Label')
29 | #for bar in ax.patches:
30 | # ax.annotate(format(bar.get_height(), ''),
31 | # (bar.get_x() + bar.get_width() / 2,
32 | # bar.get_height()), ha='center', va='center',
33 | # size=15, xytext=(0, 8),
34 | # textcoords='offset points')
35 |
36 | ################
37 | # CS top 30 Utilization comparison
38 | mal_cs_data = mal.filter(regex='grp_')
39 | #mal_cs_data = mal_cs_data.drop(['cs_len'], axis=1)
40 | #mal_cs_data = mal_cs_data.loc[:, (mal_cs_data != 0).any(axis=0)]
41 | #mal_cs_data = pd.concat([mal_cs_data, mal.malware_label], axis=1)
42 |
43 | ben_cs_data = ben.filter(regex='grp_')
44 | #ben_cs_data = ben_cs_data.drop(['cs_len'], axis=1)
45 | #ben_cs_data = ben_cs_data.loc[:, (ben_cs_data != 0).any(axis=0)]
46 | #ben_cs_data = pd.concat([ben_cs_data, ben.malware_label], axis=1)
47 |
48 | #cs_sum = pd.concat([ben_cs_data.sum(), mal_cs_data.sum()], axis=1)
49 | cs_sum = pd.DataFrame({"Benign": ben_cs_data.sum(), "Malware": mal_cs_data.sum()}, index=ben_cs_data.sum().sort_values().index)
50 | #cs_sum = cs_sum.loc[:, (cs_sum != 0).any(axis=0)]
51 | #cs_sum = pd.DataFrame({"Benign": ben_cs_data.sum().sort_values()}, index=ben_cs_data.sum().sort_values().index)
52 | #cs_sum = pd.DataFrame({"Malware": mal_cs_data.sum().sort_values()}, index=mal_cs_data.sum().sort_values().index)
53 |
54 | ###################
55 | # Length Fields
56 | #mal_len_data = mal.filter(regex='_len')
57 | #mal_len_data = mal_len_data.loc[:, (mal_len_data != 0).any(axis=0)]
58 |
59 | #ben_len_data = ben.filter(regex='_len')
60 | #ben_len_data = ben_len_data.loc[:, (ben_len_data != 0).any(axis=0)]
61 |
62 | #cs_sum = pd.concat([pd.DataFrame(ben_len_data.mean(), columns=['Benign']), pd.DataFrame(mal_len_data.mean(), columns=['Malware'])], axis=1)
63 |
64 | #################
65 | # Svr features
66 | #mal_svr_data = mal.filter(regex='handshake_')
67 | #mal_svr_data = mal[['dom_in_tranco_1m', 'dom_dga_prob', 'otx_status', 'otx_age', 'urlhaus_status','urlhaus_age']]
68 | #mal_svr_data = mal_svr_data.drop(['svr_tls_ver', 'svr_supported_ver'], axis=1)
69 | #mal_svr_data = pd.DataFrame(pd.DataFrame(MinMaxScaler().fit_transform(mal_svr_data), columns=mal_svr_data.columns).mean(), columns=['Malware'])
70 |
71 | #ben_svr_data = ben.filter(regex='handshake_')
72 | #ben_svr_data = ben[['dom_in_tranco_1m', 'dom_dga_prob', 'otx_status', 'otx_age', 'urlhaus_status','urlhaus_age']]
73 | #ben_svr_data = ben_svr_data.drop(['svr_tls_ver', 'svr_supported_ver'], axis=1)
74 | #ben_svr_data = pd.DataFrame(pd.DataFrame(MinMaxScaler().fit_transform(ben_svr_data), columns=ben_svr_data.columns).mean(), columns=['Benign'])
75 |
76 | #cs_sum = pd.concat([ben_svr_data, mal_svr_data], axis=1).sort_values(by='Malware')
77 |
78 | ################
79 | # Ports
80 | #mal_prt = mal[['src_port', 'dst_port']]
81 | #ben_prt = ben[['src_port', 'dst_port']]
82 | #ben_percent = ben_prt.dst_port.unique().shape[0] / ben_prt.dst_port.shape[0]
83 | #mal_percent = mal_prt.dst_port.unique().shape[0] / mal_prt.dst_port.shape[0]
84 | #cs_sum = pd.DataFrame({'Label': ['Benign Unique Dst Ports', 'Malware Unique Dst Ports'], 'Count': [ ben_percent, mal_percent ]})
85 | #cs_sum = pd.DataFrame(ben_prt.dst_port.value_counts()).sort_values(by='dst_port').tail(10)
86 | #cs_sum = cs_sum[cs_sum.index <= 30000]
87 | #cs_sum = pd.concat([ben_cs_data, mal_cs_data], axis=1).sort_values(by='Malware').tail(40)
88 | #mal_cs_nz = len(mal_cs_data.loc[:, (mal_cs_data != 0).any(axis=0)].columns)
89 | #ben_cs_nz = len(ben_cs_data.loc[:, (ben_cs_data != 0).any(axis=0)].columns)
90 |
91 | # Add labels above bars
92 | #cs_count = pd.DataFrame({'Label': ['Benign Cipher Suites', 'Malware Cipher Suites'], 'Count': [ben_cs_nz, mal_cs_nz]})
93 |
94 | #ax = sns.barplot(data=cs_count, y='Count', x='Label', orient='v')
95 |
96 | #ax.set_xticklabels(ax.get_xticklabels(), rotation=0)
97 | #for bar in ax.patches:
98 | # ax.annotate(format(bar.get_height(), ''),
99 | # (bar.get_x() + bar.get_width() / 2,
100 | # bar.get_height()), ha='center', va='center',
101 | # size=8, xytext=(0, 8),
102 | # textcoords='offset points')
103 | #ax.set(xlabel='Label', ylabel='Count')
104 | cs_sum.plot.barh(rot=0)
105 | plt.show()
--------------------------------------------------------------------------------
/feature-reduction-tests/reduce_features.py:
--------------------------------------------------------------------------------
1 | import pandas as pd
2 |
3 | from tensorflow.keras.layers import Input, Dense
4 | from tensorflow.keras.models import Model
5 | from tensorflow.keras import regularizers
6 | from tensorflow.random import set_seed
7 | from sklearn.preprocessing import MinMaxScaler
8 | from sklearn.decomposition import PCA
9 | from numpy.random import seed
10 | from sklearn.ensemble import RandomForestClassifier
11 |
12 | def pca_output (data, label, components=10):
13 | '''
14 | Used to generate a desired number of "Principle Components" from the input data and return the calculated
15 | output components as a pandas dataframe
16 | '''
17 | data_label = data[[label]]
18 | data = data.drop([label], axis=1)
19 | std_data = MinMaxScaler().fit_transform(data.values)
20 |
21 | # Calculate the 10 most important components
22 | pca = PCA(n_components = components)
23 | data_pca_vals = pca.fit_transform(std_data)
24 | pca_dataframe = pd.DataFrame(data = data_pca_vals)
25 | final_pca_dataframe = pd.concat([pca_dataframe, data_label], axis=1)
26 |
27 | return pd.DataFrame(final_pca_dataframe)
28 |
29 |
30 | def random_forest(data, label, estimators):
31 | '''
32 | Use a Random Forest to determine feature importance. Can also be used to return the top number of
33 | features determined by the "IF" statement of val_tuple[1] value. This is the threshold determined by
34 | the Random Forest regressor and can be determiend by generating and analyzing the bar chart.
35 |
36 | This proved to be between 65%-80% successful in classifing malicious traffic by itself, however, the SVM
37 | used in svm-testing.py was more successful due to its robust ability to deal with outliers.
38 | '''
39 | seed(1)
40 | set_seed(2)
41 | SEED = 123
42 | data_label = data.malware_label
43 | data = data.drop([label], axis=1)
44 | features = data.columns
45 |
46 | # Standardize the dataset
47 | std_data = MinMaxScaler().fit_transform(data)
48 |
49 | regressor = RandomForestClassifier(n_estimators=estimators, random_state=SEED)
50 | regressor.fit(std_data, data_label)
51 | feat_series = regressor.feature_importances_
52 |
53 | feature_list = [(feature, round(importance, 2)) for feature, importance in zip(list(features), list(feat_series))]
54 | top_x_feature_list = []
55 | for val_tuple in feature_list:
56 | if val_tuple[1] >= 0.03:
57 | top_x_feature_list.append(val_tuple[0])
58 |
59 | final_data = pd.DataFrame(MinMaxScaler().fit_transform(data[top_x_feature_list]))
60 | final_data = pd.concat([final_data, data_label], axis=1)
61 |
62 | # Return dataframe of N most important features based on feature_list importances and
63 | # val_tuple[0] measure value
64 | return final_data
65 |
66 | def autoencoded_features (data, label, feature_count):
67 | '''
68 | Attempt to analyze and interpret data using a Sparse, Stacked Autoencoder. Was not too successful,
69 | but leaving here for potential, future analysis in feature reduction in lieu of Random Forest or
70 | PCA. Read some interesting research where that was successful.
71 | '''
72 | # Balance dataset based on percentage passed to function
73 | seed(1)
74 | set_seed(2)
75 | data_label = data.malware_label
76 | data = data.drop([label], axis=1)
77 | std_data = MinMaxScaler().fit_transform(data)
78 |
79 | # Autoencoder values
80 | input_dim = data.shape[1]
81 | encoding_dim = int(input_dim / 2)
82 | hidden_dim_1 = int(encoding_dim / 2)
83 | hidden_dim_2 = int(hidden_dim_1 / 2)
84 | final_hidden_dim = feature_count
85 | learning_rate = 1e-6
86 |
87 | input_layer = Input(shape=(input_dim, ))
88 | encoder = Dense(encoding_dim, activation='relu', activity_regularizer=regularizers.l1(learning_rate))(input_layer)
89 | encoder = Dense(hidden_dim_1, activation='relu')(encoder)
90 | encoder = Dense(hidden_dim_2, activation='relu')(encoder)
91 | encoder = Dense(final_hidden_dim, activation='relu')(encoder)
92 |
93 | autoencoder_model = Model(inputs=input_layer, outputs=encoder)
94 | autoencoder_feature_output = pd.DataFrame(autoencoder_model.predict(std_data))
95 | autoencoder_feature_output = pd.concat([autoencoder_feature_output, data_label], axis=1)
96 |
97 | return autoencoder_feature_output
98 |
99 | if __name__ == '__main__':
100 | #def load_input (csv_data_file):
101 | csv_data_file = 'test_train_data-all.csv'
102 | dataset = pd.read_csv(csv_data_file)
103 | # Remove columns filled with all 0 value (these will be statistically insignifant and will cause
104 | # issues when using correlation methods of analysis)
105 | data_no_z_cols = dataset.loc[:, (dataset != 0).any(axis=0)]
106 |
107 | # NOTE TO SELF
108 | # To import as module for testing:
109 | # from importlib import import_module, reload
110 | # a = import_module('features')
111 | # Use this when you make changes
112 | # reload(a)
113 | #
114 | # Call methods using below syntax:
115 | # a.random_forest(a.dataset, 'labelname', 100, 'bar')
--------------------------------------------------------------------------------
/format-data/check_ip.py:
--------------------------------------------------------------------------------
1 | #!/usr/bin/env python
2 |
3 | # This script tells if a File, IP, Domain or URL may be malicious according to the data in a couple of
4 | # OSINT sources
5 | import IndicatorTypes
6 | import json
7 | import socket
8 | import requests
9 | import os
10 | import time
11 | import logging
12 | import csv
13 | import ipaddress
14 |
15 | from datetime import date, datetime, timedelta
16 | from dateutil.parser import parse
17 | from OTXv2 import OTXv2
18 | from tranco import Tranco
19 |
20 | # Define logger for cross appliction logging consistency
21 | logger = logging.getLogger(__name__)
22 |
23 | def getValue(results, keys):
24 | '''
25 | Get a nested key from a dict, without having to do loads of ifs
26 | '''
27 |
28 | if type(keys) is list and len(keys) > 0:
29 | if type(results) is dict:
30 | key = keys.pop(0)
31 | if key in results:
32 | return getValue(results[key], keys)
33 | else:
34 | return None
35 | else:
36 | if type(results) is list and len(results) > 0:
37 | return getValue(results[0], keys)
38 | else:
39 | return results
40 | else:
41 | return results
42 |
43 | def date_range(start_date, end_date):
44 | '''
45 | Return a list of the dates from start to end date
46 | '''
47 | try:
48 | dates = []
49 | for day in range (int ((end_date - start_date).days) + 1):
50 | dates.append(start_date + timedelta(day))
51 | except Exception as e:
52 | logging.exception("There was a problem generating dates... {}".format(e))
53 | exit(1)
54 |
55 | return dates
56 |
57 | def is_ipv4(string):
58 | try:
59 | ipaddress.IPv4Network(string)
60 | return True
61 | except ValueError:
62 | return False
63 |
64 | def ip(api_key, ip):
65 | '''
66 | Query AlienVault OTX for malicious IP checking
67 | '''
68 | alerts = {}
69 | full_url_list = []
70 | report_age = 0
71 | url_status = 0
72 |
73 | if ipaddress.ip_address(ip).is_private:
74 | alerts = {'url_status': 0, 'report_age': 0}
75 | else:
76 | # Set URLs and instantiate OTX object
77 | OTX_SERVER = 'https://otx.alienvault.com/'
78 | otx = OTXv2(api_key, server=OTX_SERVER)
79 |
80 | try:
81 | result = otx.get_indicator_details_full(IndicatorTypes.IPv4, ip)
82 | pulses = getValue(result['general'], ['pulse_info', 'pulses'])
83 | if pulses:
84 | full_url_list = getValue(result['passive_dns'], ['passive_dns'])
85 | if full_url_list:
86 | first_reported = datetime.strptime(full_url_list[-1]['first'], "%Y-%m-%dT%H:%M:%S")
87 | last_reported = datetime.strptime(full_url_list[0]['last'], "%Y-%m-%dT%H:%M:%S")
88 | report_age = int(((last_reported - first_reported).total_seconds() / 86400))
89 | else:
90 | report_age = 0
91 | url_status = 1
92 |
93 | except (RetryError):
94 | url_status = 1
95 | report_age = 0
96 | except Exception as e:
97 | logger.exception(" : You received this error with the OTX API Data... {}".format(e))
98 |
99 | # Build alerts dictionary from variables
100 | alerts = {'url_status': url_status, 'report_age': report_age}
101 |
102 | return alerts
103 |
104 | def hostname(host, ip_addr, query_two=False):
105 | '''
106 | Query urlhaus.abuse.ch for malicious URL checking (if no URL found, check IP as some malicious domains are the unresolved IP)
107 | '''
108 | alerts = {}
109 | report_age = 0
110 | url_status = 0
111 | query_results = ''
112 | __version__ = '0.0.2'
113 |
114 | # URL haus API URL
115 | urlhaus_api = "https://urlhaus-api.abuse.ch/v1/"
116 |
117 | if is_ipv4(ip_addr) and ipaddress.ip_address(ip_addr).is_private:
118 | query_two = True
119 |
120 | try:
121 | query_urlhaus = requests.post("{}host/".format(urlhaus_api), headers={"User-Agent" : "urlhaus-python-client-{}".format(__version__)}, data={"host": host})
122 | if query_urlhaus.ok:
123 | query_results = query_urlhaus.json()
124 | if query_results['query_status'] == "no_results" and not query_two:
125 | hostname(ip_addr, host, True)
126 |
127 | elif query_results['query_status'] == "no_results" and query_two:
128 | url_status = 0
129 | report_age = 0
130 |
131 | else:
132 | if not query_urlhaus.json()['urls']:
133 | url_status = 0.5
134 | report_age = 0
135 | elif query_urlhaus.json()['urls'][0]['url_status'] == 'online':
136 | first_seen = datetime.strptime(query_urlhaus.json()['firstseen'], "%Y-%m-%d %H:%M:%S UTC")
137 | last_seen = datetime.now()
138 | report_age = int(((last_seen - first_seen).total_seconds() / 86400))
139 | url_status = 1
140 | else:
141 | url_status = 0.5
142 | report_age = 0
143 |
144 | else:
145 | logger.error(" : Unable to read response as json")
146 |
147 | except Exception as e:
148 | logger.exception(" : Unable to connect to URLHaus API. Recieved the following error {} - {} - {}".format(e, host, ip_addr))
149 |
150 | # Build alerts dictionary from variables
151 | alerts = {'url_status': url_status, 'report_age': report_age}
152 | return alerts
153 |
154 | def update_csv(csv_file):
155 | '''
156 | Method to update ja3 csv file from sslbl.urlhaus.ch
157 | '''
158 | with open(csv_file, 'wb') as write_ja3_csv:
159 | csv_data = requests.get('https://sslbl.abuse.ch/blacklist/ja3_fingerprints.csv')
160 | write_ja3_csv.write(csv_data.content)
161 |
162 | def ja3_sslbl_check(fingerprint):
163 | '''
164 | Method to check JA3 database for existence of connection
165 | '''
166 | ja3_malware_check = {}
167 | csv_filename = r'C:\Users\bryan\Desktop\ja3_fingerprints.csv'
168 | days = 1
169 | report_age = 0
170 | ja3_check = 0
171 |
172 | if not os.path.exists(csv_filename):
173 | update_csv(csv_filename)
174 | else:
175 | file_modify_date = os.path.getmtime(csv_filename)
176 | file_older_than_days = ((time.time() - file_modify_date) / 3600 > 24 * days)
177 | if file_older_than_days:
178 | update_csv(csv_filename)
179 |
180 | with open(csv_filename, 'r') as ja3_csv_file:
181 | # Add csv file to variable and ignore comments
182 | read_ja3 = csv.reader(filter(lambda row: row[0] != '#', ja3_csv_file))
183 | # Locate value in csv (if exists)
184 | for ja3_fingerprint in read_ja3:
185 | if fingerprint == ja3_fingerprint[0]:
186 | ja3_check = 1
187 | first_seen = datetime.strptime(ja3_fingerprint[1], "%Y-%d-%m %H:%M:%S")
188 | last_seen = datetime.strptime(ja3_fingerprint[2], "%Y-%d-%m %H:%M:%S")
189 | report_age = int((last_seen - first_seen).total_seconds() / 86400)
190 | break
191 |
192 | ja3_malware_check = {'ja3_check': ja3_check, 'ja3_record_age': report_age}
193 |
194 | return ja3_malware_check
195 |
196 | def dns_tranco_check(cache_dir, domain_name, number_of_days):
197 | '''
198 | Analyze DNS domains for Tranco 1 million over the last 30 days and return percentage existence
199 | '''
200 | # Set variables
201 | tranco_result = float()
202 | begin_date = date.today() - timedelta(number_of_days)
203 | # Must set time delta to not consider the last 2 days. This prevents errors when the latest
204 | # records have not been released during the evaluation timeframe
205 | end_date = date.today() - timedelta(2)
206 | date_check_range = date_range(begin_date, end_date)
207 | tranco_data = Tranco(cache=True, cache_dir=cache_dir)
208 | occurence_count = 0
209 |
210 | # Iterate over date range and increase count for each occurence of the domain
211 | try:
212 | for single_day in date_check_range:
213 | tranco_1M = set(tranco_data.list(single_day.strftime("%Y-%m-%d")).top(1000000))
214 | if domain_name in tranco_1M:
215 | occurence_count += 1
216 | else:
217 | occurence_count += 0
218 | except (ValueError, AttributeError):
219 | occurence_count += 0
220 | except Exception as e:
221 | logging.exception("There was a problem in Tranco analysis... {}".format(e))
222 | #exit(1)
223 |
224 | # Convert occurence value to float for ML analysis
225 | tranco_result = occurence_count / number_of_days
226 |
227 | return tranco_result
228 |
229 | def main():
230 | '''
231 | Run some automated tests for ip and hostname methods
232 | '''
233 | api_key = os.environ.get('API_KEY')
234 | ip_addr = '4.4.4.4'
235 | host = 'google.com'
236 |
237 | print("IP Query output: {}".format(ip(api_key, ip_addr)))
238 | print("Hostname Query output: {}".format(hostname(host, ip_addr)))
239 |
240 | if __name__ == "__main__":
241 | try:
242 | exit(main())
243 | except Exception:
244 | logging.exception("Exception in main()")
245 | exit(1)
--------------------------------------------------------------------------------
/format-data/extract_data_csv.py:
--------------------------------------------------------------------------------
1 | #!/usr/bin/env python3
2 |
3 | import csv
4 | import logging
5 | import os
6 | import argparse
7 |
8 | from check_ip import ip, hostname, ja3_sslbl_check, dns_tranco_check
9 | from dgaintel import get_prob
10 | from concurrent.futures import ProcessPoolExecutor
11 | from functools import partial
12 |
13 | # Define logger for cross appliction logging consistency
14 | logger = logging.getLogger(__name__)
15 |
16 | ip_domain_dict = {}
17 |
18 | csv_header = ['dom_in_tranco_1m', 'dom_dga_prob', 'otx_status', 'otx_age', 'urlhaus_status',
19 | 'urlhaus_age', 'ja3_urlhaus_status', 'ja3_urlhaus_age', 'tls_record_type', 'client_tls_ver',
20 | 'message_len', 'handshake_type', 'handshake_version', 'handshake_len',
21 | 'cs_len', 'ext_len', 'src_port', 'dst_port', 'cs_0000', 'cs_0001', 'cs_0002',
22 | 'cs_0003', 'cs_0004', 'cs_0005', 'cs_0006', 'cs_0007', 'cs_0008','cs_0009',
23 | 'cs_000a', 'cs_000b', 'cs_000c', 'cs_000d', 'cs_000e', 'cs_000f', 'cs_0010',
24 | 'cs_0011', 'cs_0012','cs_0013', 'cs_0014', 'cs_0015', 'cs_0016', 'cs_0017', 'cs_0018',
25 | 'cs_0019', 'cs_001a', 'cs_001b', 'cs_001e', 'cs_001f', 'cs_0020', 'cs_0021', 'cs_0022',
26 | 'cs_0023', 'cs_0024', 'cs_0025', 'cs_0026', 'cs_0027', 'cs_0028', 'cs_0029',
27 | 'cs_002a', 'cs_002b', 'cs_002c', 'cs_002d', 'cs_002e', 'cs_002f', 'cs_0030',
28 | 'cs_0031', 'cs_0032', 'cs_0033', 'cs_0034', 'cs_0035', 'cs_0036', 'cs_0037',
29 | 'cs_0038', 'cs_0039', 'cs_003a', 'cs_003b', 'cs_003c', 'cs_003d', 'cs_003e',
30 | 'cs_003f', 'cs_0040', 'cs_0041', 'cs_0042', 'cs_0043', 'cs_0044', 'cs_0045',
31 | 'cs_0046', 'cs_0067', 'cs_0068', 'cs_0069', 'cs_006a', 'cs_006b', 'cs_006c',
32 | 'cs_006d', 'cs_0084', 'cs_0085', 'cs_0086', 'cs_0087', 'cs_0088', 'cs_0089',
33 | 'cs_008a', 'cs_008b', 'cs_008c', 'cs_008d', 'cs_008e', 'cs_008f', 'cs_0090',
34 | 'cs_0091', 'cs_0092', 'cs_0093', 'cs_0094', 'cs_0095', 'cs_0096', 'cs_0097',
35 | 'cs_0098', 'cs_0099', 'cs_009a', 'cs_009b', 'cs_009c', 'cs_009d', 'cs_009e',
36 | 'cs_009f', 'cs_00a0', 'cs_00a1', 'cs_00a2', 'cs_00a3', 'cs_00a4', 'cs_00a5',
37 | 'cs_00a6', 'cs_00a7', 'cs_00a8', 'cs_00a9', 'cs_00aa', 'cs_00ab', 'cs_00ac',
38 | 'cs_00ad', 'cs_00ae', 'cs_00af', 'cs_00b0', 'cs_00b1', 'cs_00b2', 'cs_00b3',
39 | 'cs_00b4', 'cs_00b5', 'cs_00b6', 'cs_00b7', 'cs_00b8', 'cs_00b9', 'cs_00ba',
40 | 'cs_00bb', 'cs_00bc', 'cs_00bd', 'cs_00be', 'cs_00bf', 'cs_00c0', 'cs_00c1',
41 | 'cs_00c2', 'cs_00c3', 'cs_00c4', 'cs_00c5', 'cs_00c6', 'cs_00c7', 'cs_00ff',
42 | 'cs_1301', 'cs_1302', 'cs_1303', 'cs_1304', 'cs_1305', 'cs_5600', 'cs_c001',
43 | 'cs_c002', 'cs_c003', 'cs_c004', 'cs_c005', 'cs_c006', 'cs_c007', 'cs_c008',
44 | 'cs_c009', 'cs_c00a', 'cs_c00b', 'cs_c00c', 'cs_c00d', 'cs_c00e', 'cs_c00f',
45 | 'cs_c010', 'cs_c011', 'cs_c012', 'cs_c013', 'cs_c014', 'cs_c015', 'cs_c016',
46 | 'cs_c017', 'cs_c018', 'cs_c019', 'cs_c01a', 'cs_c01b', 'cs_c01c', 'cs_c01d',
47 | 'cs_c01e', 'cs_c01f', 'cs_c020', 'cs_c021', 'cs_c022', 'cs_c023', 'cs_c024',
48 | 'cs_c025', 'cs_c026', 'cs_c027', 'cs_c028', 'cs_c029', 'cs_c02a', 'cs_c02b',
49 | 'cs_c02c', 'cs_c02d', 'cs_c02e', 'cs_c02f', 'cs_c030', 'cs_c031', 'cs_c032',
50 | 'cs_c033', 'cs_c034', 'cs_c035', 'cs_c036', 'cs_c037', 'cs_c038', 'cs_c039',
51 | 'cs_c03a', 'cs_c03b', 'cs_c03c', 'cs_c03d', 'cs_c03e', 'cs_c03f', 'cs_c040',
52 | 'cs_c041', 'cs_c042', 'cs_c043', 'cs_c044', 'cs_c045', 'cs_c046', 'cs_c047',
53 | 'cs_c048', 'cs_c049', 'cs_c04a', 'cs_c04b', 'cs_c04c', 'cs_c04d', 'cs_c04e',
54 | 'cs_c04f', 'cs_c050', 'cs_c051', 'cs_c052', 'cs_c053', 'cs_c054', 'cs_c055',
55 | 'cs_c056', 'cs_c057', 'cs_c058', 'cs_c059', 'cs_c05a', 'cs_c05b', 'cs_c05c',
56 | 'cs_c05d', 'cs_c05e', 'cs_c05f', 'cs_c060', 'cs_c061', 'cs_c062', 'cs_c063',
57 | 'cs_c064', 'cs_c065', 'cs_c066', 'cs_c067', 'cs_c068', 'cs_c069', 'cs_c06a',
58 | 'cs_c06b', 'cs_c06c', 'cs_c06d', 'cs_c06e', 'cs_c06f', 'cs_c070', 'cs_c071',
59 | 'cs_c072', 'cs_c073', 'cs_c074', 'cs_c075', 'cs_c076', 'cs_c077', 'cs_c078',
60 | 'cs_c079', 'cs_c07a', 'cs_c07b', 'cs_c07c', 'cs_c07d', 'cs_c07e', 'cs_c07f',
61 | 'cs_c080', 'cs_c081', 'cs_c082', 'cs_c083', 'cs_c084', 'cs_c085', 'cs_c086',
62 | 'cs_c087', 'cs_c088', 'cs_c089', 'cs_c08a', 'cs_c08b', 'cs_c08c', 'cs_c08d',
63 | 'cs_c08e', 'cs_c08f', 'cs_c090', 'cs_c091', 'cs_c092', 'cs_c093', 'cs_c094',
64 | 'cs_c095', 'cs_c096', 'cs_c097', 'cs_c098', 'cs_c099', 'cs_c09a', 'cs_c09b',
65 | 'cs_c09c', 'cs_c09d', 'cs_c09e', 'cs_c09f', 'cs_c0a0', 'cs_c0a1', 'cs_c0a2',
66 | 'cs_c0a3', 'cs_c0a4', 'cs_c0a5', 'cs_c0a6', 'cs_c0a7', 'cs_c0a8', 'cs_c0a9',
67 | 'cs_c0aa', 'cs_c0ab', 'cs_c0ac', 'cs_c0ad', 'cs_c0ae', 'cs_c0af', 'cs_c0b0',
68 | 'cs_c0b1', 'cs_c0b2', 'cs_c0b3', 'cs_c0b4', 'cs_c0b5', 'cs_c100', 'cs_c101',
69 | 'cs_c102', 'cs_c103', 'cs_c104', 'cs_c105', 'cs_c106', 'cs_cca8', 'cs_cca9',
70 | 'cs_ccaa', 'cs_ccab', 'cs_ccac', 'cs_ccad', 'cs_ccae', 'cs_d001', 'cs_d002',
71 | 'cs_d003', 'cs_d005', 'cs_unknown', 'sig_0201', 'sig_0203', 'sig_0401', 'sig_0403', 'sig_0420',
72 | 'sig_0501', 'sig_0503', 'sig_0520', 'sig_0601', 'sig_0603', 'sig_0620', 'sig_0704',
73 | 'sig_0705', 'sig_0706', 'sig_0707', 'sig_0708', 'sig_0709', 'sig_070A', 'sig_070B',
74 | 'sig_070C', 'sig_070D', 'sig_070E', 'sig_070F', 'sig_0804', 'sig_0805', 'sig_0806',
75 | 'sig_0807', 'sig_0808', 'sig_0809', 'sig_080a', 'sig_080b', 'sig_081a', 'sig_081b',
76 | 'sig_081c', 'sig_grease', 'sig_empty', 'grp_01', 'grp_02', 'grp_03', 'grp_04', 'grp_05',
77 | 'grp_06', 'grp_07', 'grp_08', 'grp_09', 'grp_10', 'grp_11', 'grp_12', 'grp_13', 'grp_14',
78 | 'grp_15', 'grp_16', 'grp_17', 'grp_18', 'grp_19', 'grp_20', 'grp_21', 'grp_22',
79 | 'grp_23', 'grp_24', 'grp_25', 'grp_26', 'grp_27', 'grp_28', 'grp_29', 'grp_30',
80 | 'grp_31', 'grp_32', 'grp_33', 'grp_34', 'grp_35', 'grp_36', 'grp_37', 'grp_38',
81 | 'grp_39', 'grp_40', 'grp_41', 'grp_256', 'grp_257', 'grp_258', 'grp_259', 'grp_260',
82 | 'grp_65281', 'grp_65282', 'grp_grease', 'pts_00', 'pts_01', 'pts_02', 'svr_ext_00', 'svr_ext_01',
83 | 'svr_ext_02','svr_ext_03', 'svr_ext_04', 'svr_ext_05', 'svr_ext_06', 'svr_ext_07', 'svr_ext_08',
84 | 'svr_ext_09', 'svr_ext_10', 'svr_ext_11', 'svr_ext_12', 'svr_ext_13', 'svr_ext_14',
85 | 'svr_ext_15', 'svr_ext_16', 'svr_ext_17', 'svr_ext_18', 'svr_ext_19', 'svr_ext_20',
86 | 'svr_ext_21', 'svr_ext_22', 'svr_ext_23', 'svr_ext_24', 'svr_ext_25', 'svr_ext_26',
87 | 'svr_ext_27', 'svr_ext_28', 'svr_ext_29', 'svr_ext_30', 'svr_ext_31', 'svr_ext_32',
88 | 'svr_ext_33', 'svr_ext_34', 'svr_ext_35', 'svr_ext_36', 'svr_ext_37', 'svr_ext_38',
89 | 'svr_ext_39', 'svr_ext_40', 'svr_ext_41', 'svr_ext_42', 'svr_ext_43', 'svr_ext_44',
90 | 'svr_ext_45', 'svr_ext_46', 'svr_ext_47', 'svr_ext_48', 'svr_ext_49', 'svr_ext_50',
91 | 'svr_ext_51', 'svr_ext_52', 'svr_ext_53', 'svr_ext_55', 'svr_ext_56', 'svr_ext_65281',
92 | 'svr_ext_unassigned', 'svr_ocsp_staple', 'svr_tls_ver', 'svr_supported_ver', 'malware_label']
93 |
94 | # Create custom logging class for exceptions
95 | class OneLineExceptionFormatter(logging.Formatter):
96 | def formatException(self, exc_info):
97 | result = super().formatException(exc_info)
98 | return repr(result)
99 |
100 | def format(self, record):
101 | result = super().format(record)
102 | if record.exc_text:
103 | result = result.replace("\n", "")
104 | return result
105 |
106 | def write_csv_file(filename, data, header=False):
107 | '''
108 | Write data to csv
109 | '''
110 | if header:
111 | try:
112 | with open(filename, "w", newline='') as outfile:
113 | write_csv = csv.DictWriter(outfile, fieldnames=csv_header)
114 | write_csv.writeheader()
115 | except Exception as e:
116 | logging.exception("There was a problem in the CSV file write process... {}".format(e))
117 | exit(1)
118 | else:
119 | try:
120 | with open(filename, "a", newline='') as outfile:
121 | write_csv = csv.DictWriter(outfile, fieldnames=csv_header)
122 | write_csv.writerow(data)
123 | except Exception as e:
124 | logging.exception("There was a problem in the CSV file write process... {}".format(e))
125 | exit(1)
126 |
127 | def correlate_data(csv_filename, tls_server_list, malware_label, API_KEY, out_dir, tls_client_entry):
128 | '''
129 | Reads in data from netcap TLS files and returns dictionary to insert into CSV file
130 | '''
131 | test_train_data = {}
132 | global ip_domain_dict
133 | ip_domain_value = ""
134 | tls_osint_list = []
135 | tranco_cache_dir = os.path.join(out_dir, '.tranco')
136 |
137 | # Pre-generate test_train_data_dict with 0 values
138 | for val in csv_header:
139 | test_train_data[val] = 0
140 |
141 | test_train_data['malware_label'] = malware_label
142 |
143 | for tls_server_data in tls_server_list:
144 | if tls_server_data['SrcIP'] == tls_client_entry['DstIP'] and tls_server_data['DstIP'] == tls_client_entry['SrcIP'] and tls_server_data['DstPort'] == tls_client_entry['SrcPort'] and tls_server_data['SrcPort'] == tls_client_entry['DstPort']:
145 | tls_server_entry = tls_server_data
146 | # Drop entries from list for search efficiency and to reduce invalid duplicates
147 | tls_server_list.remove(tls_server_entry)
148 | ip_domain_value = "{}:{}".format(tls_client_entry['DstIP'], tls_client_entry['SNI'])
149 | break
150 |
151 | # Check to see if key is in the ip_domain_dict (meaning it is an existing entry)
152 | # If it is, then skip check_ip, otherwise, we'll run it and add the values to the dict
153 | if ip_domain_value in ip_domain_dict.keys():
154 | tls_osint_list.append(ip_domain_dict[ip_domain_value][0])
155 | tls_osint_list.append(ip_domain_dict[ip_domain_value][1])
156 | tls_osint_list.append(ip_domain_dict[ip_domain_value][2])
157 | tls_osint_list.append(ip_domain_dict[ip_domain_value][3])
158 | tls_osint_list.append(ip_domain_dict[ip_domain_value][4])
159 | else:
160 | dst_ip = tls_client_entry['DstIP']
161 | sni = tls_client_entry['SNI']
162 | tls_osint_list.append(ip(API_KEY, dst_ip))
163 | tls_osint_list.append(hostname(sni, dst_ip))
164 | tls_osint_list.append(ja3_sslbl_check(tls_client_entry['Ja3']))
165 |
166 | # Tranco and dgaintel fail when domain name is empty
167 | if not sni == '':
168 | tls_osint_list.append(dns_tranco_check(tranco_cache_dir, sni, 15))
169 | tls_osint_list.append(get_prob(sni))
170 | else:
171 | tls_osint_list.append(0)
172 | tls_osint_list.append(0)
173 |
174 | ip_domain_dict[ip_domain_value] = tls_osint_list
175 |
176 | # OSINT OTX and urlhaus analysis
177 | test_train_data['otx_status'] = tls_osint_list[0]['url_status']
178 | test_train_data['otx_age'] = tls_osint_list[0]['report_age']
179 | test_train_data['urlhaus_status'] = tls_osint_list[1]['url_status']
180 | test_train_data['urlhaus_age'] = tls_osint_list[1]['report_age']
181 | test_train_data['ja3_urlhaus_status'] = tls_osint_list[2]['ja3_check']
182 | test_train_data['ja3_urlhaus_age'] = tls_osint_list[2]['ja3_record_age']
183 | test_train_data['dom_in_tranco_1m'] = tls_osint_list[3]
184 | test_train_data['dom_dga_prob'] = tls_osint_list[4]
185 |
186 | # Set TLS Client static data fields in test_train_data dict
187 | test_train_data['tls_record_type'] = tls_client_entry['Type']
188 | test_train_data['client_tls_ver'] = tls_client_entry['Version']
189 | test_train_data['message_len'] = tls_client_entry['MessageLen']
190 | test_train_data['handshake_type'] = tls_client_entry['HandshakeType']
191 | test_train_data['handshake_version'] = tls_client_entry['HandshakeVersion']
192 | test_train_data['handshake_len'] = tls_client_entry['HandshakeLen']
193 | test_train_data['cs_len'] = tls_client_entry['CipherSuiteLen']
194 | test_train_data['ext_len'] = tls_client_entry['ExtensionLen']
195 | test_train_data['src_port'] = tls_client_entry['SrcPort']
196 | test_train_data['dst_port'] = tls_client_entry['DstPort']
197 |
198 | # Set TLS Server static data fields in test_train_data dict
199 | test_train_data['svr_tls_ver'] = tls_server_entry['Version']
200 | test_train_data['svr_supported_ver'] = tls_server_entry['SupportedVersion']
201 | if tls_server_entry['OCSPStapling'] == 'false':
202 | test_train_data['svr_ocsp_staple'] = 0
203 | else:
204 | test_train_data['svr_ocsp_staple'] = 1
205 |
206 | svr_selected_group = "{:02}".format(int(tls_server_entry['SelectedGroup']))
207 | server_cs_used = "{:04x}".format(int(tls_server_entry['CipherSuite']))
208 |
209 | try:
210 | # Cipher Suites
211 | tls_client_entry['CipherSuites'] = tls_client_entry['CipherSuites'][1:-1].split('-')
212 | for cs_val in tls_client_entry['CipherSuites']:
213 | entry_hex = "{:04x}".format(int(cs_val))
214 |
215 | cs_entry = "cs_{}".format(entry_hex)
216 |
217 | if cs_entry in test_train_data:
218 | test_train_data[cs_entry] += 0.5
219 | else:
220 | test_train_data['cs_unknown'] += 0.5
221 |
222 | if entry_hex == server_cs_used:
223 | test_train_data[cs_entry] += 0.5
224 |
225 | # Signature Algorithms
226 | tls_client_entry['SignatureAlgs'] = tls_client_entry['SignatureAlgs'][1:-1].split('-')
227 |
228 | if not tls_client_entry['SignatureAlgs'][0]:
229 | tls_client_entry['SignatureAlgs'] = ['0']
230 |
231 | sig_reserved_count = 1
232 | for sig_data in tls_client_entry['SignatureAlgs']:
233 | sig_entry = "sig_{:04x}".format(int(sig_data))
234 |
235 | if sig_entry in test_train_data:
236 | test_train_data[sig_entry] = 0.5
237 | elif sig_entry == 'sig_0000':
238 | test_train_data['sig_empty'] = 0.5
239 | else:
240 | test_train_data['sig_grease'] = sig_reserved_count
241 | sig_reserved_count += 0.5
242 |
243 | # Supported Groups
244 | tls_client_entry['SupportedGroups'] = tls_client_entry['SupportedGroups'][1:-1].split('-')
245 | for grp_data in tls_client_entry['SupportedGroups']:
246 | grp_val = "{:02}".format(int(grp_data))
247 |
248 | grp_entry = "grp_{}".format(grp_val)
249 |
250 | if grp_entry in test_train_data:
251 | test_train_data[grp_entry] = 0.5
252 | else:
253 | grp_entry = 'grp_grease'
254 | test_train_data[grp_entry] += 0.5
255 |
256 | if grp_val == svr_selected_group:
257 | test_train_data[grp_entry] += 0.5
258 |
259 | # Supported Points
260 | if tls_client_entry['SupportedPoints'][0] == '(':
261 | tls_client_entry['SupportedPoints'] = tls_client_entry['SupportedPoints'][1:-1].split('-')
262 | else:
263 | tls_client_entry['SupportedPoints'] = [tls_client_entry['SupportedPoints']]
264 |
265 | for pts_data in tls_client_entry['SupportedPoints']:
266 | pts_entry = "pts_{:02}".format(int(pts_data))
267 | test_train_data[pts_entry] = 0.5
268 |
269 | # Server Extensions
270 | if tls_server_entry['Extensions'][0] == '(':
271 | tls_server_entry['Extensions'] = tls_server_entry['Extensions'][1:-1].split('-')
272 | elif not type(tls_server_entry['Extensions']) is list and not tls_server_entry['Extensions'][0] == '':
273 | tls_server_entry['Extensions'] = [tls_server_entry['Extensions'][1:]]
274 | elif tls_server_entry['Extensions'][0] == '':
275 | tls_server_entry['Extensions'] = ['0']
276 |
277 | for svr_ext_data in tls_server_entry['Extensions']:
278 | svr_ext_entry = "svr_ext_{:02}".format(int(svr_ext_data))
279 |
280 | if svr_ext_entry in test_train_data:
281 | test_train_data[svr_ext_entry] = 0.5
282 | else:
283 | test_train_data['svr_ext_unassigned'] += 0.5
284 |
285 | except Exception as e:
286 | print("The problem is with a loop... - {}".format(e))
287 |
288 | write_csv_file(csv_filename, test_train_data)
289 |
290 | def main():
291 | '''
292 | Gather and format data from the TLSClientHello and TLSServerHello CSV files generated from NetCap
293 | '''
294 | # Get a label value for known data
295 | parser = argparse.ArgumentParser(description='Get label input for data analysis')
296 | parser.add_argument('-l', '--label', action='store', dest='label', default=0, help='Is this data known malicious or not?', required=False)
297 | parser.add_argument('-o', '--outfile', action='store', dest='out_file', default='test_train_data.csv',
298 | help='Name of the output file', required=False)
299 | parser.add_argument('-c', '--client-file', action='store', dest='client_file', default='TLSClientHello.csv',
300 | help='Name of the TLS Client Hello (TLSClientHello.csv) file created with NetCap', required=False)
301 | parser.add_argument('-s', '--server-file', action='store', dest='server_file', default='TLSServerHello.csv',
302 | help='Name of the TLS Server Hello (TLSServerHello.csv) file created with NetCap', required=False)
303 | parser.add_argument('-a', '--api-key', action='store', dest='api', default='',
304 | help='API Key value required for Alienvault OTX', required=False)
305 |
306 | options = parser.parse_args()
307 |
308 | base_log_dir = os.getcwd()
309 | out_dir = r'C:\Users\bryan\Desktop'
310 | csv_filename = os.path.join(out_dir, options.out_file)
311 | tls_client_file = os.path.join(base_log_dir, options.client_file)
312 | tls_server_file = os.path.join(base_log_dir, options.server_file)
313 | tls_server_list = []
314 | malicious_label = options.label
315 | #API_KEY = os.environ.get('API_KEY')
316 | API_KEY = options.api
317 |
318 | if not os.path.exists(csv_filename):
319 | write_csv_file(csv_filename, csv_header, True)
320 |
321 | with open(tls_server_file, 'r', newline='') as tls_server_data:
322 | tls_server_csv = csv.DictReader(tls_server_data)
323 | for line in tls_server_csv:
324 | tls_server_list.append(line)
325 |
326 | with ProcessPoolExecutor() as executor:
327 | fn = partial(correlate_data, csv_filename, tls_server_list, malicious_label, API_KEY, out_dir)
328 | with open(tls_client_file, 'r', newline='') as tls_client_data:
329 | tls_client_csv = csv.DictReader(tls_client_data)
330 | executor.map(fn, tls_client_csv, timeout=86400)
331 |
332 | if __name__ == '__main__':
333 | try:
334 | exit(main())
335 | except Exception:
336 | logging.exception("Exception in main()")
337 | exit(1)
--------------------------------------------------------------------------------
/format-data/ja3_fingerprints.csv:
--------------------------------------------------------------------------------
1 | ################################################################
2 | # abuse.ch Suricata JA3 Fingerprint Blacklist (CSV) #
3 | # For Suricata 4.1.0 or newer #
4 | # Last updated: 2020-04-09 06:48:14 UTC #
5 | # #
6 | # Terms Of Use: https://sslbl.abuse.ch/blacklist/ #
7 | # For questions please contact sslbl [at] abuse.ch #
8 | ################################################################
9 | #
10 | # ja3_md5,Firstseen,Lastseen,Listingreason
11 | b386946a5a44d1ddcc843bc75336dfce,2017-07-14 18:08:15,2019-07-27 20:42:54,Dridex
12 | 8991a387e4cc841740f25d6f5139f92d,2017-07-14 19:02:03,2019-07-28 00:34:38,Adware
13 | cb98a24ee4b9134448ffb5714fd870ac,2017-07-14 19:48:28,2019-05-22 03:22:38,Dridex
14 | 1aa7bf8b97e540ca5edd75f7b8384bfa,2017-07-14 20:23:38,2019-07-28 01:38:22,TrickBot
15 | 3d89c0dfb1fa44911b8fa7523ef8dedb,2017-07-15 04:23:45,2020-12-06 17:43:38,Adware
16 | bc6c386f480ee97b9d9e52d472b772d8,2017-07-15 10:57:38,2020-12-06 09:52:14,Adware
17 | 8f52d1ce303fb4a6515836aec3cc16b1,2017-07-15 19:05:11,2019-07-27 20:00:57,TrickBot
18 | d6f04b5a910115f4b50ecec09d40a1df,2017-07-15 19:42:24,2018-10-14 08:12:51,Dridex
19 | 35c0a31c481927f022a3b530255ac080,2017-07-15 19:43:19,2020-10-31 18:36:39,Tofsee
20 | d551fafc4f40f1dec2bb45980bfa9492,2017-07-15 19:59:29,2020-11-16 13:06:20,Adware
21 | e330bca99c8a5256ae126a55c4c725c5,2017-07-15 19:59:29,2020-11-01 17:49:10,Adware
22 | b8f81673c0e1d29908346f3bab892b9b,2017-07-16 01:32:03,2018-12-17 06:08:03,Adware
23 | 83e04bc58d402f9633983cbf22724b02,2017-07-16 01:32:03,2019-04-30 15:04:48,Adware
24 | 70722097d1fe1d78d8c2164640ab6df4,2017-07-16 02:39:08,2020-12-07 11:17:16,Tofsee
25 | 9c2589e1c0e9f533a022c6205f9719e1,2017-07-16 08:37:17,2020-12-01 17:48:11,Adware
26 | 849b04bdbd1d2b983f6e8a457e0632a8,2017-07-16 08:37:17,2020-12-01 17:48:11,Adware
27 | 16efcf0e00504ddfedde13bfea997952,2017-07-16 19:45:45,2020-08-21 18:01:27,Adware
28 | 4d7a28d6f2263ed61de88ca66eb011e3,2017-07-16 21:20:29,2020-12-07 09:31:52,Tofsee
29 | 550dce18de1bb143e69d6dd9413b8355,2017-07-16 22:17:20,2018-12-21 07:04:50,Adware
30 | c50f6a8b9173676b47ba6085bd0c6cee,2017-07-16 22:38:41,2019-05-21 09:42:17,TrickBot
31 | 20dd18bdd3209ea718989030a6f93364,2017-07-18 10:22:58,2019-04-28 09:23:31,Adware
32 | 8498fe4268764dbf926a38283e9d3d8f,2017-07-18 10:22:58,2019-04-28 09:23:31,Adware
33 | 590a232d04d56409fab72e752a8a2634,2017-07-18 18:53:24,2020-10-11 20:48:33,Tofsee
34 | 51a7ad14509fd614c7bb3a50c4982b8c,2017-07-19 07:28:19,2019-07-14 11:58:32,JBifrost
35 | 96eba628dcb2b47607192ba74a3b55ba,2017-07-19 18:53:48,2020-11-13 17:37:05,Tofsee
36 | df5c30e670dba99f9270ed36060cf054,2017-07-20 17:44:07,2018-04-11 15:57:59,Tofsee
37 | 098f55e27d8c4b0a590102cbdb3a5f3a,2017-07-21 09:52:01,2019-04-08 01:09:54,Adware
38 | 46efd49abcca8ea9baa932da68fdb529,2017-07-22 14:07:36,2020-12-06 21:06:38,Adware
39 | 29085f03f8e8a03f0b399c5c7cf0b0b8,2017-07-22 14:07:36,2020-12-07 03:30:34,Adware
40 | d7150af4514b868defb854db0f62a441,2017-07-23 09:39:24,2018-07-24 01:04:58,Tofsee
41 | 03e186a7f83285e93341de478334006e,2017-07-24 18:17:14,2020-10-31 18:36:38,Tofsee
42 | 3cda52da4ade09f1f781ad2e82dcfa20,2017-07-30 18:41:36,2019-05-21 17:34:18,Quakbot
43 | b13d01846ad7a14a70bf030a16775c78,2017-08-08 07:12:49,2020-12-06 21:22:44,Adware
44 | 1543a7c46633acf71e8401baccbd0568,2017-08-08 21:32:28,2020-11-10 05:30:17,Tofsee
45 | 1d095e68489d3c535297cd8dffb06cb9,2017-08-12 19:56:28,2020-10-28 11:06:23,Tofsee
46 | 93d056782d649deb51cda44ecb714bb0,2017-08-28 12:20:47,2019-04-15 23:47:27,Adware
47 | 698e36219f3979420fa2581b21dac7ec,2017-08-28 12:20:47,2019-04-28 09:23:31,Adware
48 | 1712287800ac91b34cadd5884ce85568,2017-08-28 16:01:59,2020-12-04 20:11:58,TorrentLocker
49 | 5e573c9c9f8ba720ef9b18e9fce2e2f7,2017-08-30 13:44:56,2020-12-05 20:27:10,Adware
50 | f6fd83a21f9f3c5f9ff7b5c63bbc179d,2017-10-20 08:03:21,2018-11-06 06:42:12,Adware
51 | 92579701f145605e9edc0b01a901c6d5,2017-10-23 00:10:48,2020-12-06 01:11:23,Adware
52 | a61299f9b501adcf680b9275d79d4ac6,2017-11-04 18:03:59,2020-04-21 17:08:24,Tofsee
53 | b2b61db7b9490a60d270ccb20b462826,2017-11-14 20:12:03,2020-12-06 01:08:46,Adware
54 | 7dcce5b76c8b17472d024758970a406b,2017-11-22 12:42:46,2020-10-24 19:47:51,Tofsee
55 | 534ce2dbc413c68e908363b5df0ae5e0,2017-12-22 09:36:21,2019-07-27 15:22:33,TrickBot
56 | fb00055a1196aeea8d1bc609885ba953,2018-01-01 22:49:25,2019-04-09 06:58:58,TrickBot
57 | a50a861119aceb0ccc74902e8fddb618,2018-01-02 08:16:23,2018-07-05 02:33:08,Tofsee
58 | e7643725fcff971e3051fe0e47fc2c71,2018-01-31 08:06:13,2020-03-25 16:19:48,Tofsee
59 | 7c410ce832e848a3321432c9a82e972b,2018-01-31 20:04:25,2020-12-07 11:17:17,Tofsee
60 | da949afd9bd6df820730f8f171584a71,2018-02-03 05:19:37,2020-12-06 09:52:14,Tofsee
61 | 906004246f3ba5e755b043c057254a29,2018-03-11 08:25:38,2018-04-14 00:59:16,Tofsee
62 | fd80fa9c6120cdeea8520510f3c644ac,2018-03-11 09:34:30,2020-12-07 11:09:14,Tofsee
63 | b90bdbe961a648f0427db21aaa6ccb59,2018-03-11 10:37:43,2020-05-29 23:39:01,Tofsee
64 | 1fe4c7a3544eb27afec2adfb3a3dbf60,2018-03-11 19:23:08,2020-12-07 11:17:17,Tofsee
65 | c201b92f8b483fa388be174d6689f534,2018-03-12 13:43:52,2020-08-17 16:56:42,Gozi
66 | 9f62c4f26b90d3d757bea609e82f2eaf,2018-03-13 06:23:41,2020-11-15 11:42:42,Tofsee
67 | 1be3ecebe5aa9d3654e6e703d81f6928,2018-03-13 11:50:02,2020-11-29 16:18:00,Ransomware.Troldesh
68 | e3b2ab1f9a56f2fb4c9248f2f41631fa,2018-03-15 01:06:34,2020-12-07 11:17:17,Tofsee
69 | dff8a0aa1c904aaea76c5bf624e88333,2018-03-18 09:41:15,2020-10-27 09:50:24,Tofsee
70 | 17fd49722f8d11f3d76dce84f8e099a7,2018-03-19 23:02:27,2020-12-06 15:29:30,Tofsee
71 | 911479ac8a0813ed1241b3686ccdade9,2018-03-19 23:24:59,2020-03-30 04:09:18,Tofsee
72 | c5deb9465d47232dd48772f9c4d14679,2018-03-22 15:42:48,2020-11-26 18:49:34,Tofsee
73 | f22bdd57e3a52de86cda40da2d84e83b,2018-03-27 13:40:19,2019-01-20 14:31:39,Tofsee
74 | d18a4da84af59e1108862a39bae7c9d4,2018-04-03 00:40:51,2020-11-11 00:58:38,Tofsee
75 | 2d8794cb7b52b777bee2695e79c15760,2018-04-04 06:56:37,2020-12-06 20:44:32,Ransomware
76 | 40adfd923eb82b89d8836ba37a19bca1,2018-04-15 15:49:08,2020-12-07 00:23:55,CoinMiner
77 | 1aee0238942d453d679fc1e37a303387,2018-05-13 01:59:49,2020-12-04 09:02:41,Tofsee
78 | 2092e1fffb45d7e4a19a57f9bc5e203a,2018-05-16 21:59:36,2018-09-05 01:58:33,Adware
79 | bffa4501966196d3d6e90cee1f88fc89,2018-06-07 15:08:04,2020-03-16 00:03:44,Tofsee
80 | 807fca46d9d0cf63adf4e5e80e414bbe,2018-06-07 16:51:03,2020-12-06 19:46:37,Tofsee
81 | fb58831f892190644fe44e25bc830b45,2018-06-08 12:07:59,2019-05-31 21:02:37,Adware
82 | 0cc1e84568e471aa1d62ad4158ade6b5,2018-06-24 10:50:47,2020-11-24 14:41:00,Tofsee
83 | d2935c58fe676744fecc8614ee5356c7,2018-08-14 21:48:41,2020-12-07 09:31:53,Adwind
84 | 8916410db85077a5460817142dcbc8de,2018-08-21 12:32:28,2020-12-07 03:30:34,TrickBot
85 | c5235d3a8b9934b7fbbd204d50bc058d,2018-08-23 17:36:08,2019-10-13 05:11:09,Gootkit
86 | 57f3642b4e37e28f5cbe3020c9331b4c,2018-08-28 15:54:53,2020-12-07 04:22:05,Gozi
87 | e62a5f4d538cbf169c2af71bec2399b4,2018-08-30 15:45:40,2020-12-06 21:59:37,TrickBot
88 | 51c64c77e60f3980eea90869b68c58a8,2018-08-30 21:04:57,2020-12-07 05:41:18,Dridex
89 | 7691297bcb20a41233fd0a0baa0a3628,2018-09-17 02:50:05,2020-12-07 09:00:41,Adware
90 | 7dd50e112cd23734a310b90f6f44a7cd,2018-09-17 17:54:58,2020-12-06 23:16:28,Quakbot
91 | 52c7396a501e4fecbdfa99c5408334ac,2018-09-18 00:29:04,2019-12-03 17:24:02,Tofsee
92 | f735bbc6b69723b9df7b0e7ef27872af,2018-10-02 18:04:16,2020-12-06 01:47:38,TrickBot
93 | 49ed2ef3f1321e5f044f1e71b0e6fdd5,2018-10-02 18:04:17,2020-12-06 01:47:38,TrickBot
94 | d76ee64fb7273733cbe455ac81c292e6,2018-11-16 13:26:39,2018-11-18 19:19:36,Tofsee
95 | 8f6c918dcb585ebbea05e2cc94530e3d,2018-11-16 13:26:41,2020-05-06 15:45:21,Tofsee
96 | 34f14a69ad7009ca5863379218af17f3,2018-11-17 05:17:22,2018-12-29 01:46:46,Tofsee
97 | c2b4710c6888a5d47befe865c8e6fb19,2018-11-29 20:46:04,2020-12-06 20:41:48,Tofsee
98 | decfb48a53789ebe081b88aabb58ee34,2018-12-21 09:06:16,2020-10-08 10:41:00,Adwind
99 | 08a8a4e85b25ac42e1490bc85cfdb5ce,2019-01-30 02:48:34,2020-10-27 09:50:19,Tofsee
100 | c0220cd64849a629397a9cb68f78a0ea,2019-03-24 00:12:32,2020-12-07 11:17:17,Tofsee
101 | 7a29c223fb122ec64d10f0a159e07996,2019-06-09 22:55:29,2020-10-27 09:50:26,Tofsee
102 | 44dab16d680ef93487bc16ad23b3ffb1,2019-06-09 22:55:29,2020-10-27 09:50:25,Tofsee
103 | 70a04365be5bbd4653698bebeb43ce68,2019-07-02 06:26:56,2020-05-30 04:19:00,Tofsee
104 | d81d654effb94714a4086734fa0adad9,2019-07-16 23:29:02,2020-10-27 09:50:21,Tofsee
105 | 25d74b7b4b779eb1efd4b31d26d651c6,2019-08-03 20:15:33,2020-07-14 21:43:25,Tofsee
106 | fc2299d5b2964cd242c5a2c8c531a5f0,2019-08-09 23:56:32,2020-12-07 11:17:17,Tofsee
107 | 32926ca3e59f0413d0b98725454594f5,2019-09-12 06:56:10,2020-10-27 21:49:31,Tofsee
108 | ffefafdb86336d057eda5fdf02b3d5ce,2019-10-26 07:31:49,2020-07-25 00:14:09,Tofsee
109 | # END (98) entries
--------------------------------------------------------------------------------
/format-data/requirements.txt:
--------------------------------------------------------------------------------
1 | tranco==0.5
2 | dgaintel==2.3
3 | OTXv2==1.5.10
--------------------------------------------------------------------------------
/graph/SAVE MODEL GRAPHS HERE:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/graph/SAVE MODEL GRAPHS HERE
--------------------------------------------------------------------------------
/models/ADD TRAINED MODEL HERE:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/ADD TRAINED MODEL HERE
--------------------------------------------------------------------------------
/models/ae.h5:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/ae.h5
--------------------------------------------------------------------------------
/models/oc-svm.pkl:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/oc-svm.pkl
--------------------------------------------------------------------------------
/models/svm.pkl:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/svm.pkl
--------------------------------------------------------------------------------
/models/win-pkl-ver/oc-svm.pkl:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/win-pkl-ver/oc-svm.pkl
--------------------------------------------------------------------------------
/models/win-pkl-ver/svm.pkl:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/models/win-pkl-ver/svm.pkl
--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | sklearn
2 | numpy==1.18.5
3 | pandas==0.25.3
4 | tensorflow==2.3.2
5 | seaborn
6 | matplotlib
7 | mlxtend
--------------------------------------------------------------------------------
/test-train-data/test_train_data.7z:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/1computerguy/tls-mal-detect/d93b287529bbc53871b5c9e6623af1301b7dc920/test-train-data/test_train_data.7z
--------------------------------------------------------------------------------