Android Malware Detection via Hybrid Static-Dynamic Analysis and Metaheuristic Feature Selection

Abstract

The rapid proliferation of Android applications has significantly increased the exposure of mobile devices to malware, necessitating accurate and robust detection mechanisms. While hybrid malware analysis that combines static and dynamic features has been widely adopted for Android malware detection, the effectiveness of such systems strongly depends on the selection of discriminative features. In this study, we present a systematic optimization and comparative evaluation of metaheuristic feature selection techniques within a hybrid malware analysis framework. Specifically, three metaheuristic feature selection algorithms, namely particle swarm optimization, genetic algorithm, and bat algorithm, are employed to identify the most relevant and discriminative set of features while minimizing redundancy and boosting the accuracy. An extensive set of experiments is carried out on a self-created dataset of Android applications to assess the effectiveness of the suggested approach. To further validate its effectiveness and generalizability, experiments are also performed on a benchmark dataset of Android malware. Five machine learning and three deep learning algorithms are trained with original features and the features selected using optimization algorithms for static, dynamic, and hybrid analysis approaches. Experimental results demonstrate that metaheuristic-based feature optimization consistently improves detection accuracy compared to non-optimized hybrid feature sets. According to experimental results, a gated recurrent unit constructed with features derived from the bat optimization approach for hybrid analysis gives the highest accuracy of 99.45%.