ANDROID MALWARE CLASSIFICATION APPROACH BASED ON HOST-LEVEL ENCRYPTED TRAFFIC SHAPING

With the development of mobile terminals, smartphones have attracted a very huge number of users with their powerful functions. Among them, Android system is famous for its open-source and convenience, which occupies a large market share. But this also leads many attackers to use their malware to gain benefits quickly, which make it necessary to design a practical android malware detection approach. At present, there are not many pieces of research on detecting malware by analyzing Android malicious traffic. This paper examines the characteristics of malicious traffic on the host computer to construct a traffic fingerprint. It combines machine learning algorithms to build a practical detection approach which is also suitable for encrypted traffic. To distinguish similar fuzzy traffic, an additional layer named confusion classifier is added to help further malware classification. This paper uses a real-world dataset called CICAndMal2017 and simulates two classification scenarios: malware binary detection and malware category classification. The experimental results show that the accuracy of the malware binary detection reached 98.8% while the accuracy rate of malware category classification is 95.2%.