A Novel Approach Using Feature Fusion for Android Malware Detection and Classification

The proliferation of Android malware poses an increasingly significant risk to individuals using mobile devices. The efficacy of conventional malware detection techniques using signatures is diminishing due to malware creators' continuous evolution of evasion strategies. This study presents a novel methodology for classifying Android malware that combines information derived from conventional signature-based techniques with transfer learning-based Bidirectional Encoder Representations from Transformers model. The features that have been extracted are subsequently combined to generate a novel feature vector. The aforementioned feature vector is subsequently employed to train a machine-learning classifier to detect Android malware. To strengthen the feature set multiple ways are used and that enhances the overall performance of classification of malicious and benign applications. The effectiveness of our methodology is also assessed using an extensive collection of Android applications, wherein we attain great classification accuracy to identify malicious software. Various evaluation measures such as accuracy, recall, precision, and F-measures have been accessed that show significant results.