Reverse Engineering of Android Malware Classification Using Semi-Supervised Learning

The escalating prevalence of Android malware poses a substantial threat to online security, necessitating innovative approaches to combat this growing menace. Identifying novel and intricate malware variants remains a formidable challenge in the realm of software security. Existing malware detection methods primarily relying on static features often prove inadequate in countering sophisticated modern malware. While dynamic detection methods hold promise, they frequently fail to leverage the full spectrum of malware characteristics. This paper proposes a novel malware classification framework utilizing semi-supervised learning methods utilizing both unsupervised and supervised learning algorithms. We use the CCCS-CIC-AndMal-2020 dataset encompassing 13 prominent malware categories and 191 eminent malware families. Our results suggested that PCA+XGBoost accomplished by reducing the original dimensionality from 142 features to a more compact set of 48 features. The utilization of PCA in conjunction with the XGBoost model not only enhances the computational efficiency but also preserves the essential information required for classification.