Feature-Based Semi-Supervised Learning Approach to Android Malware Detection

Abstract

The development of signature-based methods or Machine Learning (ML) techniques on static data has dominated automated malware detection on android platforms. However, these techniques may not detect dangerous activities that only manifest during runtime. Furthermore, there is already a significant volume of unlabeled malware data available, making the production of datasets through supervised ML approach of manual labelling expensive. For anti-virus researchers, the process of malware development poses a significant engineering challenge because they lack an effective method for capturing potentially new harmful files while removing clean and well-known files. In this research, we propose a semi-supervised ML technique to detect android malware from android permissions and Application Programmer Interface (API) call logs. The ML technique is incorporated into an android application to scan the installed applications and detect the corresponding levels of maliciousness with success. The results depict the feasibility of our proposed method and application.