Ensemble Feature Selection for Android SMS Malware Detection

As Android is the most popular operating system of mobile devices, the mobile devices with the Android operating system are a common target of the attackers. To protect the devices, various solutions are presented to date. Of them, the machine learning assisted solutions are considered more effective. However, such solutions often suffer from a limited detection accuracy and high computational complexity. To this end, we propose an ensemble feature selection method for detecting Android SMS malware. We exploit six feature selection algorithms to find the most important 12 features from the traffic of various Android SMS malware. Then, various machine learning algorithms are trained to predict whether an incoming traffic is benign or SMS malware. Through extensive experiments, the gradient boost-based classifiers are found to be the most effective. The gradient boost classifiers can detects the SMS malware with the highest accuracy (93.34%) compared to the other classifiers. The proposed method achieves 3.17% performance improvement with respect to the state-of-the-art SMS malware methods. The number of features selected by the proposed method is 33% less compared to the existing methods. In addition, up to 3-times more true positive rate can be obtained by the proposed metho.