Improved Malware Detection in Android Through Ensemble Modeling
Mamta Meena, Jyoti Gajrani · 2024
In the last decade, Android has evolved into the most widely adopted platform for mobile devices and smart home integration devices. However, as its popularity continues, it has become an increasingly attractive target for malicious activities. In the era where smartphones store a myriad of sensitive and private information, often with a persistent internet connection, the escalating trend of malware attacks on Android systems severely threatens users' privacy, financial security, device integrity, and file protection. Despite numerous attempts to counteract these threats through data-driven malware detection methods, the ever-evolving cultivation of Android malware obfuscation and other evasion techniques have rendered many approaches obsolete. Confronting these challenges, the paper introduces a modern framework for Android malware detection that harnesses the power of machine learning algorithms through ensemble modeling. The proposed ensemble model integrates the upsides of four machine learning models, i.e., SVM, GBDT, Lo-gistic Regression, Random Forest, and one deep learning model, Multilayer Perceptron. Rigorous validation demonstrates the robustness of our proposed framework against evasion attacks. On the CICInvesAndMal2019 dataset having an 8115 feature set of 960 samples, our approach achieves an impressive accuracy rate with the Ensemble Model. The dataset includes malware and benign apps from various families, providing comprehensive validation for the proposed model's effectiveness. Compared to state-of-the-art, the set forth ensemble model outperformed and achieved 97.91% accuracy and 98.66% F1-score.