On-device android malware detection method based on adaptive model through transfer learning, and recording medium and apparatus for performing the same
View Patent ↗Provided is an on-device Android malware detection method based on an adaptive model through transfer learning, including: determining whether an application is malicious or unfavorable from a list of applications installed on a device; decompiling, in the device, an Android package (APK) of the application installed on the device; transmitting the determined list and the decompiled APK file to a server in order to generate a head model in the server and use the generated head model for the transfer learning with a base model; performing malware analysis in the device using a transfer learning model received from the server for an application newly installed on the device; and providing a malware analysis result to a user through the device as a result, and since the malware analysis is performed on the device, it is possible to ensure the availability and real-time performance of enabling analysis outside of a network range.
1. An on-device Android malware detection method based on an adaptive model through transfer learning, the on-device Android malware detection method comprising:
determining whether a list of applications installed on a device includes an application that is malicious or unfavorable;
decompiling, in the device, an Android package (APK) file of the application installed on the device;
transmitting the determined list and the decompiled APK file to a server, generating a head model in the server, and using the generated head model for the transfer learning with a base model;
performing malware analysis in the device using a transfer learning model received from the server for an application newly installed on the device; and
providing a malware analysis result to a user through the device,
wherein, in the determining whether the list of applications installed on the device includes an application that is malicious or unfavorable, the application is determined to be a malicious or unfavorable application when the application is selected by the user, and the application is determined to be normal and reflected in a transfer learning model when the application is not selected by the userz,
wherein the decompiling, in the device, of the APK file of the application installed on the device is run in an Android environment by removing branch processing for each operating system of APKTool.
2. The on-device Android malware detection method of claim 1 , wherein the base model is based on a convolutional neural network (CNN) using a two-dimensional (2D) matrix as input data.
3. The on-device Android malware detection method of claim 1 , wherein the base model is generated based on a permission of a manifest file or is generated based on application programming interface (API) signature information defined in the APK file.
4. The on-device Android malware detection method of claim 1 , further comprising generating a new transfer learning model by communicating with the server according to needs of the user.
5. A non-transitory computer-readable recording medium on which a computer program for performing an on-device Android malware detection method based on an adaptive model through transfer learning, wherein the on-device Android malware detection method comprises:
determining whether a list of applications installed on a device includes an application that is malicious or unfavorable;
decompiling, in the device, an Android package (APK) file of the application installed on the device;
transmitting the determined list and the decompiled APK file to a server, generating a head model in the server, and using the generated head model for the transfer learning with a base model;
performing malware analysis in the device using a transfer learning model received from the server for an application newly installed on the device; and
providing a malware analysis result to a user through the device,
wherein, in the determining whether the list of applications installed on the device includes an application that is malicious or unfavorable, the application is determined to be a malicious or unfavorable application when the application is selected by the user, and the application is determined to be normal and reflected in a transfer learning model when the application is not selected by the user,
wherein the decompiling, in the device, of the APK file of the application installed on the device is run in an Android environment by removing branch processing for each operating system of APKTool.
6. An on-device Android malware detection apparatus based on an adaptive model through transfer learning, the on-device Android malware detection apparatus comprising:
a user environment reflection unit determining whether a list of applications installed on a device includes an application that is malicious or unfavorable;
a decompilation unit decompiling, in the device, an Android package (APK) file of the application installed on the device;
a communication unit transmitting the determined list and the decompiled APK file to a server in order to generate a head model in the server and use the generated head model for the transfer learning with a base model;
an analysis unit performing malware analysis in the device using a transfer learning model received from the server for an application newly installed on the device; and
a result providing unit providing a malware analysis result to a user through the device,
wherein the user environment reflection unit determines the application to be a malicious or unfavorable application when the application is selected by the user, and the user environment reflection unit determines the application to be normal and reflects it in a transfer learning model when the application is not selected by the user,
wherein the decompilation unit decompiles in an Android environment by removing branch processing for each operating system of APKTool.
7. The on-device Android malware detection apparatus of claim 6 , wherein the base model is based on a convolutional neural network (CNN) using a two-dimensional (2D) matrix as input data.
8. The on-device Android malware detection apparatus of claim 6 , wherein the base model is generated based on a permission of a manifest file or generated based on API signature information defined in the APK file.
9. The on-device Android malware detection apparatus of claim 6 , wherein the communication unit communicates with the server to generate a new transfer learning model according to needs of the user.