Meta-SonifiedDroid: Metaheuristics for Optimizing Sonified Android Malware Detection

To mitigate the rising threat of Android malware, researchers have been actively looking for mechanisms that will enable rapid and accurate malware detection. Recently, attention has been paid to the use of audio-based features derived through the use of music information retrieval techniques. Since the exploration of these features is still in the early stages, there is a need to continue experimentation, especially with features that have yet to be used for this task. In this paper, we present the results of an ongoing investigation into the use of audio-based features for Android malware detection. In addition to extracting new audio-based features, this research aims to find the most discriminative subset of audio-based features through a comparative evaluation of Wrapper-based metaheuristic optimization algorithms on two separate datasets. First, we sonified the Android APK datasets, then extracted 191 static audio-based features from the resultant audio datasets. Fourteen different nature-inspired Wrapper-based metaheuristic optimization algorithms were evaluated for feature selection, and the selected features were then used to train the light gradient-boosting machine (LGBM) classification model. Experimental results demonstrate that the proposed approach exhibits high discriminative capabilities that can outperform other state-of-the-art techniques. The best outcome for Android malware detection was obtained using features selected by the Genetic Algorithm, which achieved 50.26% feature reduction and an improved classification accuracy of 99.72%.