Feature selection for malicious android applications using Symmetrical Uncert Attribute Eval method

Abstract

The fast growth of tablets, smartphones has led to increase the usage of mobile applications. The Android apps have more popularity, however, the applications downloaded from third-party markets could be malware that may threaten the users' privacy. Several works used techniques to detect normal apps from malicious apps based on mining requested permissions. However, there are some set of permissions that can occur in benign and malignant applications. Redundant features could reduce the detection rate and increase the false positive rate. In this paper, we have proposed feature selection methods to identify clean and malicious applications based on selecting a set combination of permission patterns using different classification algorithms such as sequential minimal optimization (SMO), decision Tree (J48) and Naive Bayes. The experimental results show that sequential minimal optimization (SMO) combining with SymmetricalUncertAttributeEval method achieved the highest accuracy rate of 0.88, with lowest false positive rate of 0.085 and highest precision of 0.910. And the findings prove that feature selection methods enhanced the result of classification.