Enhancing Sustainability in Machine Learning-based Android Malware Detection using API calls

Abstract

The number of malware such as banking Trojans, spyware, and ransomware in Android devices has been rising. In addition, the recent evolution of Android malware makes existing malware detection techniques less effective. This paper shows that existing Android malware detection techniques based on Random Forest classifiers using Application Programming Interface (API) calls as a feature set are not sustainable on a relatively long-time scale. Then, we introduce two new machine learning techniques that exhibit high sustainability. By applying the proposed techniques to 126,000 Android apps, we obtained the highest accuracy of 97,8% and an F1-score of 98.8%.