Publication

Abstract :

In the domain of mobile security, particularly in APK (Android Application Package) malware detection, ensuring robust threat identification is critical for safeguarding user devices and sensitive data. This study addresses the need for a reliable Mobile Intrusion Detection System (MIDS) by harnessing advanced machine learning techniques. We advocate for employing tested models such as Logistic Regression, Decision Tree, and XGBClassifier, alongside the top-performing Random Forest (RF) model, which achieved an impressive accuracy of ninety-eight-point six percent. To enhance model performance, we utilize feature selection techniques like Recursive Feature Elimination (RFE) and hyperparameter optimization. The datasets chosen for training and evaluation, including real-world malicious and benign APKs, yield accuracies around ninety five percent. Furthermore, diverse APK datasets, including benchmarks and real-time streams, are used for rigorous testing. By leveraging optimized models and features, our study ensures exceptional performance in detecting various forms of APK malware, thus fortifying the reliability and efficacy of MIDS in real-world scenarios.