Harshvardhan

About the Project

In this work, we proposed a lightweight malware technique that uses on-device machine learning to validate Android application packages (APK). Android applications require permissions to undertake certain tasks. Some of these permissions, known as install-time permissions, are automatically granted by the Android OS and for other permissions, known as runtime permissions, the application needs to request the user to manually allow the permissions. Malicious android applications can abuse permissions. These permissions are listed in the Manifest file of the application package. Thus the work aimed at extracting these permissions from the APK, before the user installs the application. We use the permissions as input to the malware detection models. The analysis is performed locally. We did an in-depth analysis of various machine learning techniques for performing this binary classification. The final model is successful in performing the entire analysis locally with an accuracy of 93.2%

Publication:

Talos App: On-Device Machine Learning Using TensorFlow to Detect Android Malware (link)

Harshvardhan C Takawale and Abhishek Thakur
5th International Conference on Internet of Things: Systems, Management & Security, Oct. 2018