Advanced Android Malware Detection: Leveraging Machine Learning for Zero-Day Threat Defense

Abstract

The widespread use of Android devices has made them a prime target for malware, highlighting the critical need for effective detection mechanisms to protect users and their data. This thesis introduces an innovative mixed static analysis approach that leverages machine learning technique, specifically, ensemble learning, for Android malware detection, which integrates permission analysis, opcode examination, and bytecode visualization, capitalizing on the strengths of each method. Our comprehensive approach aims to enhance detection accuracy and adaptability, effectively countering the evolving tactics of malware developers, and providing an effective Zero-Day threat defense mechanism. Extensive experiments conducted on two datasets containing malware samples from different time periods demonstrate the superior performance of our method. We achieved a remarkable accuracy of 99.82% on the dataset comprising older malware samples, showcasing our model’s robustness in handling historical threats. For the dataset containing recent malware samples, our approach achieved a high accuracy of 96.06%, significantly outperforming other methods which exhibited notable decreases in performance with newer malware. These findings underscore the effectiveness of our integrated model in providing a robust defense against a wide range of malware behaviors. This research contributes significantly to cybersecurity by proposing an advanced and flexible solution for Android malware detection, offering practical implications for enhancing mobile device security in real-world applications.