PCSD: A Tool for Android Malware Detection

Abstract

The increasing amount and diversity of malicious applications are reducing efficiency of conventional defenses and it is necessary to create novel method for detection. Consequently, we propose PCSD, a lightweight tool for detection of Android malware by extracting statistical features from applications. As the influence of individual difference, PCSD performs cluster algorithm to reduce particularity. Meanwhile, it minimizes effect of random cluster by selecting cluster, which has minimum volatility on size per cluster, for improving detection accuracy. In our work, we collect statistical features from 5,553 malicious applications and 3,000 benign applications and build train model for detecting on the basis of different machine learning algorithms, like Bayesian ridge, Random forests, etc. Our results show that accuracy is 99.02% and AUC (Area Under Curve) is 99.51% in experiment. These results demonstrate the efficacy of PCSD to distinguish malicious and benign android applications.