Permission-Based Feature Scaling Method for Lightweight Android Malware Detection

Abstract

Android system has gained the highest market share due to its openness and portability in the mobile ecosystem. Whereas, users are suffering from serious security issues these years. Many malicious Android applications are released in popular app stores and it’s hard to distinguish them. Most of the current malware detection researches are focus on collecting features as much as possible for better performance instead of mining information inside the simple features. Scaling is an effective means to improve classification results while scaling the features for a large bundle of apps remains a challenging work. In this paper, we propose a malware detection method based on Machine Learning (ML) using permission usage analysis to cope with the rapid increase in the number of Android malware, named Permission Feature Selection (PFS). PFS uses Android permission as a classification feature with high utilization. The method greatly shortens the time cost in detection without reducing the detection accuracy and makes it possible to scan large-scale samples in a short time. Besides, various experiments were designed on real-world datasets to verify the reliability of the method. The results of the evaluation show that the proposed method performed better than other feature scaling methods with 91.2% accuracy and the average time cost reduced to less than 2 s.

This work was supported in by the National Key Research and Development Program of China-the Key Technologies for High Security Mobile Terminals (Grant No.2017YFB0801903), and the Youth Star project of the Institute of Information Engineering, CAS (No. Y8YS016104).