Enhancing android mobile security through machine learning-based malware detection using behavioral system features

Currently, with the explosive surge of Android applications, it has become much harder to preserve security for mobile devices since malicious applications still advance and spread by more advanced evasion tactics. Signature-based malware detection approaches are no longer effective for such evolutionary threats. In this paper, a Malware Detection Dataset (MDD) dataset used to integrate system calls and binder frequencies as feature vectors of traces to enhance Android mobile security by a machine learning-based malware detection framework. The proposed methodology consists of a systematic data pre-processing feature scaling, class distribution analysis strategy, and two deep learning architectures based on dense neural networks developed and evaluated. The first one is used as an initial architecture, and the second utilizes a broader architecture to enhance the generalization and classification performance. Experimental results indicate that the deep learning methodology has a good performance in identifying benign and malicious applications with high levels of accuracy, precision, recall, and F1-score. Detailed comparisons show that with the better structure, this model has remarkably better malware detection performance than the previous one. The work shows the promise of two deep learning-based models in Android malware analysis automation and demonstrates a scalable approach to improve real-time mobile security.