├── .gitignore ├── .gitmodules ├── Features_for_Malware_Detection.pdf ├── LICENSE ├── Prelim_Results_Malware_Detection.pdf ├── README.md ├── combine_features.py ├── config-template.ini ├── extract_apk.sh ├── extract_apks_parallel.sh ├── find_top_features.py ├── get_dates.py ├── get_samples.py ├── match_features.py ├── parse_disassembled.py ├── parse_maline_output.py ├── parse_ssdeep.py ├── parse_xml.py ├── plot_data.py ├── possibly_helpful_papers.txt ├── requirements.txt ├── run_trials.sh ├── sklearn_forest.py ├── sklearn_svm.py ├── sklearn_tree.py ├── sort_malicious.py └── tensorflow_learn.py /.gitignore: -------------------------------------------------------------------------------- 1 | config.ini 2 | all_apks/ 3 | malicious_apk/ 4 | benign_apk/ 5 | invalid_andrototal_responses 6 | *.json 7 | *.csv 8 | *.png 9 | *.apk 10 | results_* 11 | -------------------------------------------------------------------------------- /.gitmodules: -------------------------------------------------------------------------------- 1 | [submodule "tools"] 2 | path = tools 3 | url = https://bitbucket.org/andrototal/tools/ 4 | [submodule "paper"] 5 | path = paper 6 | url = https://git.overleaf.com/5092766fjzhxz 7 | -------------------------------------------------------------------------------- /Features_for_Malware_Detection.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/mwleeds/android-malware-analysis/84ff131cfdf14c627aa48ae405d3fe09a266f2e1/Features_for_Malware_Detection.pdf -------------------------------------------------------------------------------- /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 | {one line to give the program's name and a brief idea of what it does.} 635 | Copyright (C) {year} {name of author} 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 | {project} Copyright (C) {year} {fullname} 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 | -------------------------------------------------------------------------------- /Prelim_Results_Malware_Detection.pdf: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/mwleeds/android-malware-analysis/84ff131cfdf14c627aa48ae405d3fe09a266f2e1/Prelim_Results_Malware_Detection.pdf -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Getting an API Key 2 | AndroTotal has simplified the process for getting an API Key. Login/Create an Account at http://andrototal.org/ and you will then be able to view your profile settings. There is an API Tab which contains your key. 3 | 4 | This repository contains a set of scripts to automate the process of 5 | gathering data from malware samples, training a machine learning model 6 | on that data, and plotting its classification accuracy. 7 | 8 | 0. Make a copy of config-template.ini called config.ini and edit it. 9 | 10 | 1. Ensure that the "tools" subdirectory has been initialized ("`$ git submodule update --init tools`") 11 | 12 | 2. Either use `get_samples.py` to download samples or copy them into "all_apks" from another source. 13 | If you're using `get_samples.py`, you can monitor it in another shell by running `watch "ls -l *.apk | wc -l"` 14 | 15 | 3. `sort_malicious.py` uses andrototal.org to sort them into "malicious_apk" and "benign_apk" folders. 16 | You can monitor it in another shell by running `watch "ls -l benign_apk/*.apk | wc -l && ls -l malicious_apk/*.apk | wc -l"` 17 | 18 | 4. `extract_apks_parallel.sh` unpacks the .apk files into folders and processes some of the data therein. 19 | You can monitor it in another shell by running `watch "wc -l benign_apk/valid_apks.txt; wc -l malicious_apk/valid_apks.txt"` 20 | 21 | 5. Run one of the following scripts to generate feature vectors: 22 | * `parse_xml.py` for permissions. "app_permission_vectors.json" is generated 23 | * `parse_maline_output.py` for syscalls. "app_syscall_vectors.json" is generated. You will have to run [maline](https://github.com/soarlab/maline) first for this to work. 24 | * `parse_disassembled.py` for API calls. "app_method_vectors.json" is generated 25 | * `parse_ssdeep.py` for fuzzy hashes. "app_hash_vectors.json" is generated. You will have to run [ssdeep](http://ssdeep.sourceforge.net/) first for this to work. 26 | * `combine_features.py` for a combination of the top weighted features. "app_feature_vectors.json" is generated. This only works if you've previously trained a network on the specified features, and the feature weights files are named appropriately. 27 | 28 | 6. Run `$ run_trials.sh app_feature_vectors.json` (or whichever json you want) which runs the `tensorflow_learn.py` script (where the ML happens) a number of times and puts the results into a folder. It also runs `plot_data.py` and `match_features.py` to create a plot and create a list of top weighted features, respectively. 29 | 30 | 7. Change the parameters or input data and repeat step 6. It should be non-destructive so you can compare the results of different runs. 31 | 32 | Note: If you want to use a SVM instead of a neural network, use `sklearn_svm.py` in place of `tensorflow_learn.py`. You can also use `sklearn_tree.py` to use a decision tree. 33 | -------------------------------------------------------------------------------- /combine_features.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads JSON files for different features (permissions, system calls, etc.) 5 | with data on a set of apps, and reads the features' weights from a trained model, and 6 | makes a dataset with the most heavily weighted features of each type. 7 | 8 | The output data format is as follows: 9 | {"features": ["ANDROID.PERMISSION.READ_PHONE_STATE", "java/security/Signature",...], 10 | "apps": {"999eca2457729e371355aea5faa38e14.apk": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | from configparser import ConfigParser 14 | import json 15 | 16 | __author__='mwleeds' 17 | 18 | def main(): 19 | config = ConfigParser() 20 | config.read('config.ini') 21 | FEATURES = config.get('AMA', 'FEATURES').split(',') 22 | TOP_N_FEATURES = config.getint('AMA', 'TOP_N_FEATURES') 23 | INCLUDE_DATES = config.getboolean('AMA', 'INCLUDE_DATES') 24 | 25 | all_features = [] # list of strings naming each feature used in the combined dataset 26 | app_feature_map = {} # mapping from android app names to lists of features 27 | app_malicious_map = {} # mapping from android app names to 1 or 0 for malware or goodware 28 | for feature in FEATURES: 29 | with open(feature + '_weights.json') as weights: 30 | feature_weights = json.load(weights) 31 | print('Found ' + str(len(feature_weights)) + ' sets of weights for ' + feature) 32 | # no need to look at benign weights; they're complementary 33 | malicious_weights = [weight[0] for weight in feature_weights] 34 | malicious_indices = sorted(range(len(malicious_weights)), key=lambda k: malicious_weights[k], reverse=True) 35 | with open('app_' + feature + '_vectors.json') as vectors: 36 | feature_data = json.load(vectors) 37 | feature_names = feature_data['features'] 38 | print('Selecting ' + str(TOP_N_FEATURES) + ' top features of ' + str(len(feature_names))) 39 | for i in range(min(int(len(malicious_indices) / 2), int(TOP_N_FEATURES / 2))): 40 | index = malicious_indices[i] 41 | all_features.append(feature_names[index]) 42 | for i in range(min(int(len(malicious_indices) / 2), int(TOP_N_FEATURES / 2))): 43 | index = malicious_indices[-i] 44 | all_features.append(feature_names[index]) 45 | # The date feature has equal numbers of apps in each range to avoid it 46 | # being used as a feature directly, so only use those apps 47 | if INCLUDE_DATES: 48 | with open('app_date_vectors.json') as vectors: 49 | feature_data = json.load(vectors) 50 | date_buckets = feature_data['features'] 51 | all_features += date_buckets 52 | date_apps = feature_data['apps'] 53 | for app in date_apps: 54 | if app not in app_malicious_map: 55 | app_malicious_map[app] = date_apps[app]['malicious'] 56 | if app not in app_feature_map: 57 | app_feature_map[app] = [] 58 | for bucket in date_buckets: 59 | index = date_buckets.index(bucket) 60 | if date_apps[app]['vector'][index] == 1: 61 | app_feature_map[app].append(bucket) 62 | for feature in FEATURES: 63 | with open('app_' + feature + '_vectors.json') as vectors: 64 | feature_data = json.load(vectors) 65 | feature_names = feature_data['features'] 66 | feature_apps = feature_data['apps'] 67 | print('Found ' + str(len(feature_apps)) + ' apps for ' + feature) 68 | for app in feature_apps: 69 | if INCLUDE_DATES and app not in date_apps: 70 | continue 71 | if app not in app_malicious_map: 72 | app_malicious_map[app] = feature_apps[app]['malicious'] 73 | if app not in app_feature_map: 74 | app_feature_map[app] = [] 75 | for feature_name in all_features: 76 | if feature_name in feature_names: 77 | index = feature_names.index(feature_name) 78 | if feature_apps[app]['vector'][index] == 1: 79 | app_feature_map[app].append(feature_name) 80 | all_apps = {} # mapping combining app_feature_map and app_malicious_map using bits 81 | for app_name in app_feature_map: 82 | bit_vector = [1 if p in app_feature_map[app_name] else 0 for p in all_features] 83 | all_apps[app_name] = {'vector': bit_vector, 'malicious': app_malicious_map[app_name]} 84 | with open('app_feature_vectors.json', 'w') as outfile: 85 | json.dump({'features': all_features, 'apps': all_apps}, outfile) 86 | print('Wrote data on ' + str(len(all_features)) + ' features and ' + str(len(all_apps)) + ' apps to a file.') 87 | 88 | if __name__=='__main__': 89 | main() 90 | -------------------------------------------------------------------------------- /config-template.ini: -------------------------------------------------------------------------------- 1 | [AMA] 2 | API_KEY = your_api_key 3 | START_DATE = 20160401:0000 4 | END_DATE = 20161001:0000 5 | LEARNING_RATE = 0.01 6 | NUM_CHUNKS = 20 7 | FEATURES = permission,syscall,method 8 | TOP_N_FEATURES = 30 9 | SHUFFLE_CHUNKS = True 10 | DECAY_RATE = 0.975 11 | NUM_DATE_BUCKETS = 20 12 | INCLUDE_DATES = False 13 | -------------------------------------------------------------------------------- /extract_apk.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | # 4 | # This script is called by extract_apks_paprellel.sh with an apk filename, 5 | # and does several things: 6 | # 1. The apk is extracted into a folder 7 | # 2. The AndroidManifest.xml files are converted from binary to ASCII 8 | # 3. The classes.dex file is converted to classes-dex2jar.jar 9 | # 4. The classes in the jar file are printed to classes.list 10 | # 5. The jar file is disassembled to disassembled.code 11 | # 12 | # Dependencies: unzip, java, javap, xmllint, dex2jar 13 | # 14 | 15 | if [ -z "$1" ]; then 16 | echo "Usage: extract_apk.sh /path/to/apk" 17 | exit 1 18 | fi 19 | 20 | apk="$1" 21 | folder="${apk::-4}" 22 | mkdir -p "$folder" 23 | echo "Extracting $apk" 24 | unzip -u -d "$folder" "$apk" 1>/dev/null 25 | if [ $(ls -l "$folder" | wc -l) -gt 1 ]; then 26 | echo `basename $apk` >> valid_apks.txt 27 | cd "$folder" 28 | if [ -e "AndroidManifest.xml" ] && [ ! -e "AndroidManifest_ascii.xml" ]; then 29 | echo "Converting $folder/AndroidManifest.xml to ASCII" 30 | cp AndroidManifest.xml AndroidManifest_bin.xml 31 | java -jar ~/AXMLPrinter2.jar AndroidManifest_bin.xml > AndroidManifest_ascii.xml 32 | echo "Cleaning up $folder/AndroidManifest.xml" 33 | xmllint AndroidManifest_ascii.xml > AndroidManifest.xml 34 | if [ $? -ne 0 ]; then 35 | echo `basename $apk` >> ../malformed_xml.txt 36 | fi 37 | fi 38 | if [ -e "classes.dex" ] && [ ! -e "classes.list" ]; then 39 | echo "Converting $folder/classes.dex to jar format" 40 | d2j-dex2jar.sh classes.dex 41 | if [ $? -ne 0 ]; then 42 | echo `basename $apk` >> ../malformed_dex.txt 43 | else 44 | jar -tf classes-dex2jar.jar > classes.list 45 | javap -c -classpath classes-dex2jar.jar $(jar -tf classes-dex2jar.jar | grep "class$" | sed s/\.class$//) > disassembled.code 46 | fi 47 | fi 48 | cd .. 49 | fi 50 | -------------------------------------------------------------------------------- /extract_apks_parallel.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | # 4 | # This script calls the extract_apk.sh script many times in parallel 5 | # for each of the sorted applications to unpack apk files and parse 6 | # their contents. 7 | # 8 | 9 | original_dir=$(pwd) 10 | for dir in malicious_apk benign_apk; do 11 | echo "Entering $dir" 12 | cd $dir 13 | cpus=$(ls -d /sys/devices/system/cpu/cpu[[:digit:]]* | wc -w) 14 | cpus=$(expr $cpus - 2) 15 | find `pwd` -type f -name "*.apk" | xargs --max-args=1 --max-procs=$cpus ../extract_apk.sh 16 | cat valid_apks.txt | sort | uniq > valid_apks.txt 17 | cat malformed_xml.txt | sort | uniq > malformed_xml.txt 18 | cat malformed_dex.txt | sort | uniq > malformed_dex.txt 19 | cd $original_dir 20 | done 21 | -------------------------------------------------------------------------------- /find_top_features.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script prints each feature in the given JSON file along with 5 | how many apps use it, and how many of each type (malicious or benign). 6 | """ 7 | 8 | import sys 9 | import json 10 | 11 | def main(): 12 | with open(sys.argv[1]) as f: 13 | j = json.load(f) 14 | all_features = j['features'] 15 | num_apps_each = {} 16 | for app in j['apps']: 17 | for i,bit in enumerate(j['apps'][app]['vector']): 18 | if bit == 1: 19 | if j['apps'][app]['malicious'] == [1,0]: 20 | if all_features[i] in num_apps_each: 21 | num_apps_each[all_features[i]][0] += 1 22 | else: 23 | num_apps_each[all_features[i]] = [1,0] 24 | else: 25 | if all_features[i] in num_apps_each: 26 | num_apps_each[all_features[i]][1] += 1 27 | else: 28 | num_apps_each[all_features[i]] = [0,1] 29 | # This sorts by number of malicious apps using it, but can easily be changed 30 | for feature in sorted(num_apps_each.items(), key=lambda item: item[1][0], reverse=True): 31 | print('{} was used by {} apps ({} malicious and {} benign)'.format(feature[0], str(feature[1][0] + feature[1][1]), str(feature[1][0]), str(feature[1][1]))) 32 | 33 | if __name__=='__main__': 34 | if len(sys.argv) == 1: 35 | print('Usage: python3 {} '.format(sys.argv[0])) 36 | sys.exit(1) 37 | main() 38 | -------------------------------------------------------------------------------- /get_dates.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the resources.arsc files in the malicous_apk 5 | and benign_apk folders and copies the modified file dates into a 6 | JSON file for later analysis 7 | 8 | The output data format is as follows: 9 | {"features": ["1222000000_to_1222111111", ...], 10 | "apps": {"999eca2457729e371355aea5faa38e14.apk": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | import os 14 | import json 15 | import glob 16 | import time 17 | import random 18 | from configparser import ConfigParser 19 | 20 | __author__='mkkeffeler' 21 | 22 | def main(): 23 | config = ConfigParser() 24 | config.read('config.ini') 25 | NUM_DATE_BUCKETS = config.getint('AMA', 'NUM_DATE_BUCKETS') 26 | 27 | date_buckets = [] # list of strings naming each date range used in the dataset 28 | app_date_map = {} # mapping from android app names to lists of dates 29 | app_malicious_map = {} # mapping from android app names to 1 or 0 for malware or goodware 30 | apps_per_bucket = {} # number of apps in each date range 31 | relevant_buckets = [] # subset of buckets that will be used (based on NUM_DATE_BUCKETS) 32 | all_apps = {} # mapping combining app_date_map and app_malicious_map using bits 33 | apps_found_per_bucket = {} # number of apps added to all_apps for each bucket 34 | root_dir = os.getcwd() 35 | 36 | num_apps_before = 0 37 | num_apps_after = 0 38 | num_file_not_found = 0 39 | fnfe = open('file_not_found_error', 'w') 40 | for i, directory in enumerate(['benign_apk', 'malicious_apk']): 41 | os.chdir(directory) 42 | for filename in glob.glob('*.apk'): 43 | #print('Processing ' + filename) 44 | try: 45 | os.chdir(filename[:-4]) 46 | if os.path.exists('classes.dex'): 47 | mtime = os.stat('classes.dex') 48 | else: 49 | mtime = os.stat('resources.arsc') 50 | if mtime.st_mtime < 1222000000: 51 | num_apps_before += 1 52 | if mtime.st_mtime > time.time(): 53 | num_apps_after += 1 54 | except FileNotFoundError: 55 | num_file_not_found += 1 56 | fnfe.write(filename + '\n') 57 | os.chdir(os.path.join(root_dir, directory)) 58 | continue 59 | app_date_map[filename] = int(mtime.st_mtime) 60 | app_name = filename 61 | # make a one-hot bit vector of length 2. 1st bit set if malicious, otherwise 2nd bit 62 | app_malicious_map[app_name] = [1,0] if i else [0,1] 63 | os.chdir(os.pardir) 64 | os.chdir(root_dir) 65 | fnfe.close() 66 | 67 | # Android was released Sept. 23, 2008 68 | startdate = 1222000000 69 | secondsinmonth = 60 * 60 * 24 * 28 70 | enddate = startdate + secondsinmonth 71 | while True: 72 | date_buckets.append(str(startdate)+"_to_"+str(enddate)) 73 | # Apps can't have been made in the future 74 | if(enddate >= time.time()): 75 | break 76 | startdate = enddate 77 | enddate = startdate + secondsinmonth 78 | 79 | # Count the number of apps per date range so we can ensure there's an equal number in each 80 | for app_name in app_date_map: 81 | for bucket in date_buckets: 82 | mtime = app_date_map[app_name] 83 | startdate = int(bucket.split("_to_")[0]) 84 | enddate = int(bucket.split("_to_")[1]) 85 | if (startdate <= mtime) and (mtime < enddate): 86 | if bucket not in apps_per_bucket: 87 | apps_per_bucket[bucket] = app_malicious_map[app_name] 88 | else: 89 | if app_malicious_map[app_name][0] == 1: 90 | apps_per_bucket[bucket][0] += 1 91 | else: 92 | apps_per_bucket[bucket][1] += 1 93 | break 94 | with open('apps_per_bucket.json', 'w') as f: 95 | json.dump(apps_per_bucket, f) 96 | relevant_buckets = sorted(apps_per_bucket, key=lambda bucket: min(apps_per_bucket[bucket]), reverse=True) 97 | if len(relevant_buckets) > NUM_DATE_BUCKETS: 98 | relevant_buckets = relevant_buckets[:NUM_DATE_BUCKETS] 99 | apps_per_bucket_limit = min(apps_per_bucket[relevant_buckets[-1]]) # number of apps of each type (benign/malicious) of each bucket 100 | 101 | # Now add apps_per_bucket_limit apps from each bucket to all_apps 102 | for bucket in relevant_buckets: 103 | apps_found_per_bucket[bucket] = [0,0] 104 | for app_name in app_date_map: 105 | date_vector = [] 106 | in_relevant_bucket = False 107 | this_bucket = '' 108 | for bucket in relevant_buckets: 109 | mtime = app_date_map[app_name] 110 | startdate = int(bucket.split("_to_")[0]) 111 | enddate = int(bucket.split("_to_")[1]) 112 | if (startdate <= mtime) and (mtime < enddate): 113 | date_vector.append(1) 114 | in_relevant_bucket = True 115 | this_bucket = bucket 116 | else: 117 | date_vector.append(0) 118 | malicious = (app_malicious_map[app_name] == [1,0]) 119 | if in_relevant_bucket and apps_found_per_bucket[this_bucket][0 if malicious else 1] < apps_per_bucket_limit: 120 | apps_found_per_bucket[this_bucket][0 if malicious else 1] += 1 121 | all_apps[app_name] = {'vector': date_vector, 'malicious': app_malicious_map[app_name]} 122 | with open('app_date_vectors.json', 'w') as outfile: 123 | json.dump({'features': relevant_buckets, 'apps': all_apps}, outfile) 124 | print('Wrote data on ' + str(len(relevant_buckets)) + ' date buckets and ' + str(len(all_apps)) + ' apps to a file.') 125 | print('{} apps were before Android began and {} were after today'.format(num_apps_before, num_apps_after)) 126 | print('{} classes.dex files were not found'.format(num_file_not_found)) 127 | 128 | if __name__=='__main__': 129 | main() 130 | -------------------------------------------------------------------------------- /get_samples.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | # This script downloads Android malware (and goodware) from andrototal.org 4 | 5 | import subprocess 6 | import os 7 | import glob 8 | from configparser import ConfigParser 9 | 10 | def main(): 11 | config = ConfigParser() 12 | config.read('config.ini') 13 | API_KEY = config.get('AMA', 'API_KEY') 14 | START_DATE = config.get('AMA', 'START_DATE') 15 | END_DATE = config.get('AMA', 'END_DATE') 16 | subprocess.check_call('./tools/samples_cli.py getbydate -at-key {} {} {}'.format(API_KEY, START_DATE, END_DATE), shell=True) 17 | for apk in glob.glob('*.apk'): 18 | os.rename(apk, 'all_apks/' + apk) 19 | 20 | if __name__=='__main__': 21 | main() 22 | -------------------------------------------------------------------------------- /match_features.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the JSON file passed as the first command line argument to 5 | get the names of features and feature_weights.json (written by tensorflow_learn.py) 6 | to get the feature weights from a trained ML model. It then matches the feature 7 | weights to the human-readable names for them, and prints them out, sorted by the weights. 8 | So the first feature listed is the one most indicative of maliciousness, and the first 9 | one in the second list is the one most indicative of benign (the lists are just mirror 10 | images of each other). 11 | """ 12 | 13 | import json 14 | import sys 15 | 16 | def main(): 17 | with open(sys.argv[1]) as vectors: 18 | # Dataset of feature names that were used in the model 19 | feature_names = json.load(vectors)['features'] 20 | 21 | with open('feature_weights.json') as weights: 22 | # Tensorflow model calculated weights for every feature 23 | feature_weights = json.load(weights) 24 | 25 | # Separate malicous and benign weights 26 | malicious_weights = [weight[0] for weight in feature_weights] 27 | benign_weights = [weight[1] for weight in feature_weights] 28 | 29 | # Sort weights in descending order 30 | malicious_indices=sorted(range(len(malicious_weights)), key=lambda k: malicious_weights[k], reverse=True) 31 | benign_indices=sorted(range(len(benign_weights)), key=lambda k: benign_weights[k], reverse=True) 32 | 33 | # Prints the rank of each feature, its weight, and the feature name 34 | print('MALICIOUS FEATURE RANKINGS:\n') 35 | for i,x in enumerate(malicious_indices): 36 | print ('{}. {} {}'.format(i, feature_names[x], malicious_weights[x])) 37 | 38 | print ('\n\n\n\n\nBENIGN FEATURE RANKINGS:\n') 39 | for i,x in enumerate(benign_indices): 40 | print ('{}. {} {}'.format(i, feature_names[x], benign_weights[x])) 41 | 42 | if __name__=='__main__': 43 | main() 44 | -------------------------------------------------------------------------------- /parse_disassembled.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the disassembled.code files in the malicous_apk 5 | and benign_apk folders and copies the Android and Java methods called 6 | into a JSON file for later analysis. 7 | 8 | The output data format is as follows: 9 | {"features": ["java/lang/String.length", ...], 10 | "apps": {"999eca2457729e371355aea5faa38e14.apk": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | import os 14 | import json 15 | import glob 16 | 17 | __author__='mwleeds' 18 | 19 | def main(): 20 | all_methods = [] # list of strings naming each method used in the dataset 21 | app_method_map = {} # mapping from android app names to lists of methods 22 | app_malicious_map = {} # mapping from android app names to 1 or 0 for malware or goodware 23 | root_dir = os.getcwd() 24 | for i, directory in enumerate(['benign_apk', 'malicious_apk']): 25 | os.chdir(directory) 26 | category_root_dir = os.getcwd() 27 | for filename in glob.glob('*.apk'): 28 | try: 29 | print('Processing ' + filename) 30 | os.chdir(filename[:-4]) 31 | with open('disassembled.code') as disassembled_code: 32 | app_name = filename 33 | # make a one-hot bit vector of length 2. 1st bit set if malicious, otherwise 2nd bit 34 | app_malicious_map[app_name] = [1,0] if i else [0,1] 35 | # parse the file and record any interesting methods 36 | app_method_map[app_name] = [] 37 | for line in disassembled_code.readlines(): 38 | try: 39 | method = line.split('// Method ')[1].split(':')[0] 40 | #if not method.startswith('java') and not method.startswith('android'): 41 | if not method.startswith('java'): 42 | continue 43 | # Comment the below line to use methods rather than classes 44 | method = method.split('.')[0] 45 | # the method is probably obfuscated; ignore it 46 | if len(method.split('/')[-1]) < 4 or len(method.split('/')[-2]) == 1: 47 | continue 48 | if method not in all_methods: 49 | all_methods.append(method) 50 | if method not in app_method_map[app_name]: 51 | app_method_map[app_name].append(method) 52 | except IndexError: 53 | continue 54 | except FileNotFoundError as e: 55 | print(e) 56 | finally: 57 | os.chdir(category_root_dir) 58 | os.chdir(root_dir) 59 | all_apps = {} # mapping combining app_methods_map and app_malicious_map using bits 60 | for app_name in app_method_map: 61 | bit_vector = [1 if m in app_method_map[app_name] else 0 for m in all_methods] 62 | all_apps[app_name] = {'vector': bit_vector, 'malicious': app_malicious_map[app_name]} 63 | with open('app_method_vectors.json', 'w') as outfile: 64 | json.dump({'features': all_methods, 'apps': all_apps}, outfile) 65 | print('Wrote data on ' + str(len(all_methods)) + ' methods and ' + str(len(all_apps)) + ' apps to a file.') 66 | 67 | if __name__=='__main__': 68 | main() 69 | -------------------------------------------------------------------------------- /parse_maline_output.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the .freq files written by maline, converts them into bit 5 | vectors denoting which system calls were used and copies that information 6 | into a JSON file for later analysis. 7 | 8 | The output data format is as follows: 9 | {"features": ["android.permission.RECEIVE_BOOT_COMPLETED", ...], 10 | "apps": {"": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | import glob 14 | import os 15 | import json 16 | from configparser import ConfigParser 17 | 18 | __author__='mwleeds' 19 | 20 | def main(): 21 | config = ConfigParser() 22 | config.read('config.ini') 23 | MALINE_DIR = config.get('AMA', 'MALINE_DIR') 24 | MALINE_DIR = os.path.expanduser(MALINE_DIR) 25 | 26 | all_apps = {} # mapping combining app_syscall_map and app_malicious_map using bits 27 | all_syscalls = [] # list of strings naming each syscall for the architecture 28 | root_dir = os.getcwd() 29 | freq_files = glob.glob('all_freq/*.freq') 30 | os.chdir(MALINE_DIR) 31 | with open('data/i386-syscall.txt') as f: 32 | all_syscalls = [line.strip() for line in f.readlines()] 33 | os.chdir(root_dir) 34 | for filename in freq_files: 35 | print('Processing ' + filename) 36 | apk_name = filename.split('-')[1] 37 | malicious = os.path.isfile(os.getcwd() + '/malicious_apk/' + apk_name + '.apk') 38 | with open(filename) as f: 39 | frequencies = [freq for freq in f.readlines()[1].split(' ') if len(freq) > 0] 40 | assert(len(frequencies) == len(all_syscalls)) 41 | frequency_bits = [1 if int(f) > 0 else 0 for f in frequencies] 42 | all_apps[apk_name + '.apk'] = {'vector': frequency_bits, 'malicious': [1,0] if malicious else [0,1]} 43 | with open('app_syscall_vectors.json', 'w') as outfile: 44 | json.dump({'features': all_syscalls, 'apps': all_apps}, outfile) 45 | print('Wrote data on ' + str(len(all_syscalls)) + ' syscalls and ' + str(len(all_apps)) + ' apps to a file.') 46 | 47 | if __name__=='__main__': 48 | main() 49 | -------------------------------------------------------------------------------- /parse_ssdeep.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the CSV output from ssdeep in the malicious_apk 5 | and benign_apk folders and writes similarity scores for each sample 6 | and classifications to a JSON file for later analysis 7 | 8 | The output data format is as follows: 9 | {"features": ["similarity_limit_0", "similarity_limit_0.2", ...], 10 | "apps": {"999eca2457729e371355aea5faa38e14.apk": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | import os 14 | import json 15 | import glob 16 | import random 17 | import numpy 18 | import ssdeep 19 | 20 | __author__='mwleeds' 21 | 22 | def main(): 23 | all_hashes = {'malicious': [], 'benign': []} 24 | app_malicious_map = {} # mapping from android app names to 1 or 0 for malware or goodware 25 | similarity_buckets = ['similarity_limit_0', 'similarity_limit_0.2', 'similarity_limit_0.4', 'similarity_limit_0.6', 'similarity_limit_0.8', 'similarity_limit_1.0'] 26 | root_dir = os.getcwd() 27 | for i, directory in enumerate(['benign_apk', 'malicious_apk']): 28 | os.chdir(directory) 29 | with open(directory.split('_')[0] + '_apk_ssdeep.csv') as hashes: 30 | for j, line in enumerate(hashes): 31 | if j == 0: continue 32 | b64hash = line.split(',')[0] 33 | app_name = line.split(',')[-1].split('/')[-1][:-2] 34 | app_malicious_map[app_name] = [1,0] if i else [0,1] 35 | all_hashes['malicious' if i else 'benign'].append((app_name, b64hash)) 36 | os.chdir(root_dir) 37 | all_apps = {} # mapping from each app to its similarity score and classification 38 | num_zero = {} 39 | num_each = {} 40 | for category in all_hashes: 41 | num_zero[category] = 0 42 | num_each[category] = 0 43 | for app_and_hash in all_hashes[category]: 44 | similarity_scores = [] 45 | this_score = app_and_hash[1] 46 | for i in range(1000): 47 | other_score = random.choice(all_hashes[category])[1] 48 | similarity_scores.append(ssdeep.compare(this_score, other_score)) 49 | score = numpy.mean(similarity_scores) 50 | num_each[category] += 1 51 | if score == 0: num_zero[category] += 1 52 | bit_vector = [] 53 | last_limit = -0.01 54 | for limit in similarity_buckets: 55 | float_limit = float(limit.split('_')[-1]) 56 | if score <= float_limit and score > last_limit: 57 | bit_vector.append(1) 58 | else: 59 | bit_vector.append(0) 60 | last_limit = float_limit 61 | if not any(bit_vector): # score > 1 62 | bit_vector[-1] = 1 63 | all_apps[app_and_hash[0]] = {'vector': bit_vector, 'malicious': app_malicious_map[app_and_hash[0]]} 64 | with open('app_hash_vectors.json', 'w') as outfile: 65 | json.dump({'features': similarity_buckets, 'apps': all_apps}, outfile) 66 | print('{} of {} malicious apps and {} of {} benign apps had zero similarity found'.format(num_zero['malicious'], num_each['malicious'], num_zero['benign'], num_zero['benign'])) 67 | print('Wrote data on ' + str(len(all_apps)) + ' apps to a file.') 68 | 69 | if __name__=='__main__': 70 | main() 71 | -------------------------------------------------------------------------------- /parse_xml.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads the AndroidManifest.xml files in the malicous_apk 5 | and benign_apk folders and copies the requested permissions into a 6 | JSON file for later analysis 7 | 8 | The output data format is as follows: 9 | {"features": ["ANDROID.PERMISSION.RECEIVE_BOOT_COMPLETED", ...], 10 | "apps": {"999eca2457729e371355aea5faa38e14.apk": {"vector": [0,0,0,1], "malicious": [0,1]}, ...}} 11 | """ 12 | 13 | import os 14 | from defusedxml import ElementTree 15 | import json 16 | import glob 17 | 18 | __author__='mwleeds' 19 | 20 | def main(): 21 | all_permissions = [] # list of strings naming each permission used in the dataset 22 | app_permission_map = {} # mapping from android app names to lists of permissions 23 | app_malicious_map = {} # mapping from android app names to 1 or 0 for malware or goodware 24 | root_dir = os.getcwd() 25 | for i, directory in enumerate(['benign_apk', 'malicious_apk']): 26 | os.chdir(directory) 27 | category_root_dir = os.getcwd() 28 | for filename in glob.glob('*.apk'): 29 | print('Processing ' + filename) 30 | try: 31 | os.chdir(filename[:-4]) 32 | with open('AndroidManifest.xml') as xml_file: 33 | et = ElementTree.parse(xml_file) 34 | except (ElementTree.ParseError, UnicodeDecodeError, FileNotFoundError): 35 | print('Parsing error encountered for ' + filename) 36 | os.chdir(category_root_dir) 37 | continue 38 | app_name = filename 39 | # make a one-hot bit vector of length 2. 1st bit set if malicious, otherwise 2nd bit 40 | app_malicious_map[app_name] = [1,0] if i else [0,1] 41 | permissions = et.getroot().findall('./uses-permission') 42 | app_permission_map[app_name] = [] 43 | for permission in permissions: 44 | try: 45 | permission_name = permission.attrib['{http://schemas.android.com/apk/res/android}name'].upper() 46 | if not permission_name.startswith('ANDROID.PERMISSION'): continue # ignore custom permissions 47 | if permission_name not in all_permissions: all_permissions.append(permission_name) 48 | app_permission_map[app_name].append(permission_name) 49 | except KeyError: 50 | pass 51 | os.chdir(os.pardir) 52 | os.chdir(root_dir) 53 | all_apps = {} # mapping combining app_permission_map and app_malicious_map using bits 54 | for app_name in app_permission_map: 55 | bit_vector = [1 if p in app_permission_map[app_name] else 0 for p in all_permissions] 56 | all_apps[app_name] = {'vector': bit_vector, 'malicious': app_malicious_map[app_name]} 57 | with open('app_permission_vectors.json', 'w') as outfile: 58 | json.dump({'features': all_permissions, 'apps': all_apps}, outfile) 59 | print('Wrote data on ' + str(len(all_permissions)) + ' permissions and ' + str(len(all_apps)) + ' apps to a file.') 60 | 61 | if __name__=='__main__': 62 | main() 63 | -------------------------------------------------------------------------------- /plot_data.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | # This script plots the data produced by run_trials.sh 4 | 5 | import matplotlib.pyplot as plt 6 | from matplotlib.axes import Axes 7 | import numpy 8 | 9 | CSV_FILE = 'results.csv' 10 | 11 | def main(): 12 | # read the data from the CSV files 13 | data = numpy.genfromtxt(CSV_FILE, delimiter=',', names=True) 14 | 15 | # plot training steps vs classification accuracy 16 | plt.plot(data['n'], data['false_positive'], 'r.', markersize=20) 17 | plt.plot(data['n'], data['false_negative'], 'g.', markersize=20) 18 | plt.plot(data['n'], data['accuracy'], 'b.', markersize=20) 19 | plt.title('Number of Training Steps vs Classification Accuracy \n and False Negative/Positive Rates') 20 | plt.xlabel('Number of Training Steps') 21 | plt.ylabel('Classification Accuracy') 22 | plt.grid(True) 23 | plt.savefig('training_steps_vs_accuracy.png') 24 | 25 | if __name__=='__main__': 26 | main() 27 | -------------------------------------------------------------------------------- /possibly_helpful_papers.txt: -------------------------------------------------------------------------------- 1 | http://scholarworks.sjsu.edu/cgi/viewcontent.cgi?article=1488&context=etd_projects 2 | http://ieeexplore.ieee.org/xpls/icp.jsp?arnumber=5696292 3 | -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | matplotlib 2 | pydotplus 3 | sklearn 4 | numpy 5 | ssdeep 6 | -------------------------------------------------------------------------------- /run_trials.sh: -------------------------------------------------------------------------------- 1 | #!/bin/bash 2 | 3 | set -e 4 | 5 | if [ ! -e "$1" ]; then 6 | echo "Usage: $0 " 7 | exit 1 8 | fi 9 | 10 | rm -f results.csv 11 | echo "n,false_positive,false_negative,accuracy" >> results.csv 12 | for n in 20 40 60 80 100 120 140 160 180 200; do 13 | m=10 14 | false_positive_sum=0 15 | false_negative_sum=0 16 | accuracy_sum=0 17 | for i in `seq 1 $m`; do 18 | output=`python3 tensorflow_learn.py $1 $n 2>/dev/null` 19 | output_arr=(${output//$'\n'/ }) 20 | echo "n: $n false_positive: ${output_arr[0]} false_negative: ${output_arr[1]} accuracy: ${output_arr[2]}" 21 | false_positive_sum=$(echo "scale=2;$false_positive_sum + ${output_arr[0]}" | bc) 22 | false_negative_sum=$(echo "scale=2;$false_negative_sum + ${output_arr[1]}" | bc) 23 | accuracy_sum=$(echo "scale=2;$accuracy_sum + ${output_arr[2]}" | bc) 24 | NUMBER_MALICIOUS="${output_arr[3]}" 25 | NUMBER_BENIGN="${output_arr[4]}" 26 | done 27 | false_positive_avg=$(echo "scale=2;$false_positive_sum/$m" | bc) 28 | false_negative_avg=$(echo "scale=2;$false_negative_sum/$m" | bc) 29 | accuracy_avg=$(echo "scale=2;$accuracy_sum/$m" | bc) 30 | echo "$n,$false_positive_avg,$false_negative_avg,$accuracy_avg" >> results.csv 31 | done 32 | 33 | python3 plot_data.py 34 | python3 match_features.py $1 >> top_features.txt 35 | 36 | rm -f RUN_INFO 37 | grep "LEARNING_RATE" config.ini >> RUN_INFO 38 | grep "NUM_CHUNKS" config.ini >> RUN_INFO 39 | grep "SHUFFLE_CHUNKS" config.ini >> RUN_INFO 40 | grep "DECAY_RATE" config.ini >> RUN_INFO 41 | echo "NUMBER_MALICIOUS = $NUMBER_MALICIOUS" >> RUN_INFO 42 | echo "NUMBER_BENIGN = $NUMBER_BENIGN" >> RUN_INFO 43 | 44 | RUN_FOLDER="results_`date --iso-8601=seconds`" 45 | mkdir "$RUN_FOLDER" 46 | mv results.csv RUN_INFO feature_weights.json training_steps_vs_accuracy.png top_features.txt "$RUN_FOLDER" 47 | cp $1 "$RUN_FOLDER" 48 | -------------------------------------------------------------------------------- /sklearn_forest.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | import os 4 | import json 5 | import sys 6 | import sklearn 7 | from sklearn import cross_validation, grid_search, ensemble 8 | from sklearn.metrics import confusion_matrix, classification_report 9 | from sklearn.externals import joblib 10 | 11 | def train_forest_classifer(features, labels, model_output_path): 12 | """ 13 | train_forest_classifer will train a RandomForestClassifier 14 | 15 | features: 2D array of each input feature for each sample 16 | labels: array of string labels classifying each sample 17 | model_output_path: path for storing the trained forest model 18 | """ 19 | # save 20% of data for performance evaluation 20 | X_train, X_test, y_train, y_test = cross_validation.train_test_split(features, labels, test_size=0.2) 21 | 22 | param = [ 23 | { 24 | "max_depth": [None, 10, 100, 1000, 10000], 25 | "n_estimators": [1, 10, 100] 26 | } 27 | ] 28 | 29 | forest = ensemble.RandomForestClassifier(random_state=0) 30 | 31 | # 10-fold cross validation, use 4 thread as each fold and each parameter set can be train in parallel 32 | clf = grid_search.GridSearchCV(forest, param, 33 | cv=10, n_jobs=20, verbose=3) 34 | 35 | clf.fit(X_train, y_train) 36 | 37 | if os.path.exists(model_output_path): 38 | joblib.dump(clf.best_estimator_, model_output_path) 39 | else: 40 | print("Cannot save trained forest model to {0}.".format(model_output_path)) 41 | 42 | print("\nBest parameters set:") 43 | print(clf.best_params_) 44 | 45 | y_predict=clf.predict(X_test) 46 | 47 | labels=sorted(list(set(labels))) 48 | print("\nConfusion matrix:") 49 | print("Labels: {0}\n".format(",".join(labels))) 50 | print(confusion_matrix(y_test, y_predict, labels=labels)) 51 | 52 | print("\nClassification report:") 53 | print(classification_report(y_test, y_predict)) 54 | 55 | def main(): 56 | # load the feature data from a file 57 | with open(sys.argv[1]) as infile: 58 | dataset = json.load(infile) 59 | app_names = list(dataset['apps'].keys()) 60 | feature_vectors = [dataset['apps'][app]['vector'] for app in app_names] 61 | labels = ['1' if dataset['apps'][app]['malicious'] == [1,0] else '0' for app in app_names] 62 | train_forest_classifer(feature_vectors, labels, 'model.out') 63 | 64 | if __name__=='__main__': 65 | main() 66 | -------------------------------------------------------------------------------- /sklearn_svm.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | import os 4 | import json 5 | import sys 6 | import sklearn 7 | from sklearn import cross_validation, grid_search 8 | from sklearn.metrics import confusion_matrix, classification_report 9 | from sklearn.svm import SVC 10 | from sklearn.externals import joblib 11 | 12 | # Code credit to https://code.oursky.com/tensorflow-svm-image-classifications-engine/ 13 | 14 | def train_svm_classifer(features, labels, model_output_path): 15 | """ 16 | train_svm_classifer will train a SVM, saved the trained and SVM model and 17 | report the classification performance 18 | 19 | features: 2D array of each input feature for each sample 20 | labels: array of string labels classifying each sample 21 | model_output_path: path for storing the trained svm model 22 | """ 23 | # save 20% of data for performance evaluation 24 | X_train, X_test, y_train, y_test = cross_validation.train_test_split(features, labels, test_size=0.2) 25 | 26 | param = [ 27 | { 28 | "kernel": ["linear"], 29 | "C": [1, 10, 100, 1000] 30 | }, 31 | { 32 | "kernel": ["rbf"], 33 | "C": [1, 10, 100, 1000], 34 | "gamma": [1e-2, 1e-3, 1e-4, 1e-5] 35 | } 36 | ] 37 | 38 | # request probability estimation 39 | svm = SVC(probability=True) 40 | 41 | # 10-fold cross validation, use 4 thread as each fold and each parameter set can be train in parallel 42 | clf = grid_search.GridSearchCV(svm, param, 43 | cv=10, n_jobs=20, verbose=3) 44 | 45 | clf.fit(X_train, y_train) 46 | 47 | if os.path.exists(model_output_path): 48 | joblib.dump(clf.best_estimator_, model_output_path) 49 | else: 50 | print("Cannot save trained svm model to {0}.".format(model_output_path)) 51 | 52 | print("\nBest parameters set:") 53 | print(clf.best_params_) 54 | 55 | y_predict=clf.predict(X_test) 56 | 57 | labels=sorted(list(set(labels))) 58 | print("\nConfusion matrix:") 59 | print("Labels: {0}\n".format(",".join(labels))) 60 | print(confusion_matrix(y_test, y_predict, labels=labels)) 61 | 62 | print("\nClassification report:") 63 | print(classification_report(y_test, y_predict)) 64 | 65 | def main(): 66 | # load the feature data from a file 67 | with open(sys.argv[1]) as infile: 68 | dataset = json.load(infile) 69 | app_names = list(dataset['apps'].keys()) 70 | feature_vectors = [dataset['apps'][app]['vector'] for app in app_names] 71 | labels = ['1' if dataset['apps'][app]['malicious'] == [1,0] else '0' for app in app_names] 72 | train_svm_classifer(feature_vectors, labels, 'model.out') 73 | 74 | if __name__=='__main__': 75 | main() 76 | -------------------------------------------------------------------------------- /sklearn_tree.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | import os 4 | import json 5 | import sys 6 | import sklearn 7 | import pydotplus 8 | from sklearn import cross_validation, grid_search, tree 9 | from sklearn.metrics import confusion_matrix, classification_report 10 | from sklearn.externals import joblib 11 | 12 | def train_tree_classifer(features, labels, model_output_path): 13 | """ 14 | train_tree_classifer will train a DecisionTree and write it out to a pdf file 15 | 16 | features: 2D array of each input feature for each sample 17 | labels: array of string labels classifying each sample 18 | model_output_path: path for storing the trained tree model 19 | """ 20 | # save 20% of data for performance evaluation 21 | X_train, X_test, y_train, y_test = cross_validation.train_test_split(features, labels, test_size=0.2) 22 | 23 | param = [ 24 | { 25 | "max_depth": [None, 10, 100, 1000, 10000] 26 | } 27 | ] 28 | 29 | dtree = tree.DecisionTreeClassifier(random_state=0) 30 | 31 | # 10-fold cross validation, use 4 thread as each fold and each parameter set can be train in parallel 32 | clf = grid_search.GridSearchCV(dtree, param, 33 | cv=10, n_jobs=20, verbose=3) 34 | 35 | clf.fit(X_train, y_train) 36 | 37 | if os.path.exists(model_output_path): 38 | joblib.dump(clf.best_estimator_, model_output_path) 39 | else: 40 | print("Cannot save trained tree model to {0}.".format(model_output_path)) 41 | 42 | dot_data = tree.export_graphviz(clf.best_estimator_, out_file=None) 43 | graph = pydotplus.graph_from_dot_data(dot_data) 44 | graph.write_pdf('best_tree.pdf') 45 | 46 | print("\nBest parameters set:") 47 | print(clf.best_params_) 48 | 49 | y_predict=clf.predict(X_test) 50 | 51 | labels=sorted(list(set(labels))) 52 | print("\nConfusion matrix:") 53 | print("Labels: {0}\n".format(",".join(labels))) 54 | print(confusion_matrix(y_test, y_predict, labels=labels)) 55 | 56 | print("\nClassification report:") 57 | print(classification_report(y_test, y_predict)) 58 | 59 | def main(): 60 | # load the feature data from a file 61 | with open(sys.argv[1]) as infile: 62 | dataset = json.load(infile) 63 | app_names = list(dataset['apps'].keys()) 64 | feature_vectors = [dataset['apps'][app]['vector'] for app in app_names] 65 | labels = ['1' if dataset['apps'][app]['malicious'] == [1,0] else '0' for app in app_names] 66 | train_tree_classifer(feature_vectors, labels, 'model.out') 67 | 68 | if __name__=='__main__': 69 | main() 70 | -------------------------------------------------------------------------------- /sort_malicious.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | # This script looks at the .apk files in the 'all_apks' folder, 4 | # submits their hashes to andrototal.org to see if they're malicious, 5 | # and sorts them into folders based on that ('malicious_apk' and 'benign_apk') 6 | # (make sure those folders exist before running this) 7 | 8 | import subprocess 9 | import os 10 | import json 11 | import glob 12 | from configparser import ConfigParser 13 | 14 | def main(): 15 | config = ConfigParser() 16 | config.read('config.ini') 17 | API_KEY = config.get('AMA', 'API_KEY') 18 | for apk in glob.glob('all_apks/*.apk'): 19 | if not os.path.isfile(apk[:-4] + '_andrototal.json'): 20 | print('Checking ' + apk) 21 | try: 22 | analysis = subprocess.check_output('./tools/andrototal_cli.py analysis -at-key {} {}'.format(API_KEY, apk.split('/')[1][:-4]), shell=True).decode('utf-8') 23 | except subprocess.CalledProcessError as e: 24 | print(str(e)) 25 | continue 26 | with open(apk[:-4] + '_andrototal.json', 'w') as out_file: 27 | out_file.write(analysis) 28 | try: 29 | with open(apk[:-4] + '_andrototal.json') as json_file: 30 | analysis = json.load(json_file) 31 | if type(analysis) == str: raise ValueError 32 | except ValueError: 33 | with open('invalid_andrototal_responses', 'a') as out_file: 34 | out_file.write(apk.split('/')[1] + '\n') 35 | continue 36 | if all([test['result'] == 'NO_THREAT_FOUND' for test in analysis['tests']]): 37 | os.rename(apk, 'benign_apk/' + apk.split('/')[1]) 38 | else: 39 | os.rename(apk, 'malicious_apk/' + apk.split('/')[1]) 40 | 41 | if __name__=='__main__': 42 | main() 43 | -------------------------------------------------------------------------------- /tensorflow_learn.py: -------------------------------------------------------------------------------- 1 | #!/usr/bin/python3 2 | 3 | """ 4 | This script reads app_permission_vectors.json (written by parse_xml.py) and 5 | feeds the data into a tensorflow "neural network" to try to learn from it. 6 | """ 7 | 8 | import tensorflow as tf 9 | import numpy as np 10 | import json 11 | import math 12 | import random 13 | import sys 14 | from configparser import ConfigParser 15 | 16 | __author__='mwleeds' 17 | 18 | def main(): 19 | config = ConfigParser() 20 | config.read('config.ini') 21 | LEARNING_RATE = config.getfloat('AMA', 'LEARNING_RATE') 22 | NUM_CHUNKS = config.getint('AMA', 'NUM_CHUNKS') 23 | SHUFFLE_CHUNKS = config.getboolean('AMA', 'SHUFFLE_CHUNKS') 24 | DECAY_RATE = config.getfloat('AMA', 'DECAY_RATE') 25 | 26 | # load the data from a file 27 | with open(sys.argv[1]) as infile: 28 | dataset = json.load(infile) 29 | 30 | # placeholder for any number of bit vectors 31 | x = tf.placeholder(tf.float32, [None, len(dataset['features'])]) 32 | 33 | # weights for each synapse 34 | W = tf.Variable(tf.zeros([len(dataset['features']), 2])) 35 | 36 | b = tf.Variable(tf.zeros([2])) 37 | 38 | # results 39 | y = tf.nn.softmax(tf.matmul(x, W) + b) 40 | 41 | # placeholder for correct answers 42 | y_ = tf.placeholder(tf.float32, [None, 2]) 43 | 44 | cross_entropy = -tf.reduce_sum(y_*tf.log(y)) 45 | init = tf.initialize_all_variables() 46 | 47 | sess = tf.Session() 48 | 49 | sess.run(init) 50 | 51 | #TODO make a clean interface for this 52 | #TODO make sure the training sample contains reasonable proportions of benign/malicious 53 | malicious_app_names = [app for app in dataset['apps'] if dataset['apps'][app]['malicious'] == [1,0]] 54 | benign_app_names = [app for app in dataset['apps'] if dataset['apps'][app]['malicious'] == [0,1]] 55 | # break up the data into chunks for training and testing 56 | malicious_app_name_chunks = list(chunks(malicious_app_names, math.floor(len(dataset['apps']) / NUM_CHUNKS))) 57 | benign_app_name_chunks = list(chunks(benign_app_names, math.floor(len(dataset['apps']) / NUM_CHUNKS))) 58 | if SHUFFLE_CHUNKS: 59 | random.shuffle(malicious_app_name_chunks) 60 | random.shuffle(benign_app_name_chunks) 61 | 62 | # the first chunk of each will be used for testing (and the rest for training) 63 | for i in range(int(sys.argv[2])): 64 | # decayed Learning Rate increases accuracy by 3-5% 65 | decayed_learning_rate = tf.train.exponential_decay(LEARNING_RATE, i, 1, DECAY_RATE, staircase=False) 66 | train_step = tf.train.GradientDescentOptimizer(decayed_learning_rate).minimize(cross_entropy) 67 | 68 | j = random.randrange(1, len(malicious_app_name_chunks)) 69 | k = random.randrange(1, len(benign_app_name_chunks)) 70 | app_names_chunk = malicious_app_name_chunks[j] + benign_app_name_chunks[k] 71 | batch_xs = [dataset['apps'][app]['vector'] for app in app_names_chunk] 72 | batch_ys = [dataset['apps'][app]['malicious'] for app in app_names_chunk] 73 | sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys}) 74 | 75 | feature_weights=sess.run([W])[0] 76 | # store feature weights that were calculated. Used by match_features.py 77 | with open ('feature_weights.json','w') as f: 78 | json.dump(feature_weights.tolist(),f) 79 | 80 | app_names_chunk = malicious_app_name_chunks[0] + benign_app_name_chunks[0] 81 | test_xs = [dataset['apps'][app]['vector'] for app in app_names_chunk] 82 | test_ys = [dataset['apps'][app]['malicious'] for app in app_names_chunk] 83 | 84 | prediction_diff = tf.subtract(tf.argmax(y,1), tf.argmax(y_,1)) 85 | # calculate how often benign apps were thought to be malicious 86 | false_positive = tf.equal(prediction_diff, tf.constant(-1,shape=[len(test_ys)],dtype=tf.int64)) 87 | # calculate how often malicious apps were thought to be benign 88 | false_negative = tf.equal(prediction_diff, tf.constant(1,shape=[len(test_ys)],dtype=tf.int64)) 89 | # recalculate prediction accuracy 90 | correct_prediction = tf.equal(prediction_diff, tf.constant(0,shape=[len(test_ys)],dtype=tf.int64)) 91 | 92 | false_positive_rate = tf.reduce_mean(tf.cast(false_positive, tf.float32)) 93 | false_negative_rate = tf.reduce_mean(tf.cast(false_negative, tf.float32)) 94 | accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) 95 | 96 | print(sess.run(false_positive_rate, feed_dict={x: test_xs, y_: test_ys})) 97 | print(sess.run(false_negative_rate, feed_dict={x: test_xs, y_: test_ys})) 98 | print(sess.run(accuracy, feed_dict={x: test_xs, y_: test_ys})) 99 | print(str(len(malicious_app_names))) 100 | print(str(len(benign_app_names))) 101 | 102 | def chunks(l, n): 103 | """Yield successive n-sized chunks from l.""" 104 | for i in range(0, len(l), n): 105 | yield l[i:i+n] 106 | 107 | if __name__=='__main__': 108 | main() 109 | --------------------------------------------------------------------------------