Dynamic Permissions based Android Malware Detection using Machine Learning Techniques

Android is by far the most widely used mobile phone operating system around. However, Android based applications are highly vulnerable to various types of malware attacks attributed to their open nature and high popularity in the market. The fault lies in the underneath permission model of Android applications. These applications need a number of sensitive permissions during their installation and runtime, which enables possible security breaches by malware. The contributions of this paper are twofold: 1) We extract a set of 123 dynamic permissions from 11000 Android applications in a largest publicly available dataset till date; 2) We evaluate a number of machine learning classi fi cation techniques including Naive Bayes (NB), Decision Tree (J48), Random Forest (RF), Simple Logistic (SL), and k-star on the newly designed dataset for detecting malicious Android applications. The experimental results indicate that although the malware classi fi cation accuracy of RF, J48, and SL are comparable, SL performs marginally better than the other techniques.