Android Malware Detection Based on Stacking and Multi-feature Fusion

Abstract

AbstractThe rapid development of Mobile Internet technology has caused an increasing number of Internet users. However, this technology has its advantages and disadvantages, and malware such as information leakage, Trojan horses, and push advertisements are hidden in smart terminals. Smartphones are now important tools for people to communicate with each other. Android dominates the current smartphone operating system. To effectively detect malware, reducing the detection rate and improving its detection efficiency have always been historical and challenging topics. Focusing on the problems of slow detection rate, low efficiency of malware detection as well as unsatisfactory detection rate of a single kind of feature, a framework of Android malware detection combined with multiple features, MS-MalDetect, is proposed. It mainly consists of three parts: multi-feature fusion which includes permissions, opcodes, application APIs, and hardware information, feature selection which combines Information Gain and Chi-square test to reduce the dimension of features, and Stacking which constructs Machine Learning classifiers that integrates multiple models to detect malware. Experiments show the proposed MS-MalDetect achieves 96.55% and 98.56% detection accuracies in different datasets and gets better performance than the compared related works.KeywordsMalware detectionMulti-feature fusionFeature selectionStacking