Integrating CNN and XGBoost with Synthetic Samples for Advanced Android Malware Detection
Pragat Gadilohar, Deepak Singh Tomar, Vasudev Dehalwar, Yogesh Kumar Sharma · 2024
The ubiquitous presence of Android smartphones exposes users to an ever-expanding arsenal of malware threats. Existing detection methods often struggle with false positives and limited adaptability to emerging threats. This paper presents a novel hybrid deep learning approach for Android malware detection, achieving a remarkable 98.66% accuracy while minimizing false positives. By combining Convolutional Neural Networks (CNNs) with XGBoost and leveraging Generative Adversarial Networks (GANs) for data augmentation, this method demonstrates scalability and robustness across varying dataset sizes. This approach significantly enhances user privacy, device security, and app integrity, contributing to the ongoing battle against evolving Android malware threats. Utilizing a balanced dataset of 3,000 benign and malicious APK files, our method employs feature importance techniques to ensure model interpretability and relevance. The proposed approach showcases consistent performance with 1000, 2000, and 3000 APK files, highlighting its efficacy in safeguarding Android users against an ever-expanding array of malware threats.