Android Malware Detection Using Deep Learning Techniques
D R Janardhana, H V Nithin, B R Sandhya, Aakash D Bhatia · 2024
With the rapid increase in Android devices and applications, securing these platforms has become critical. Android malware presents serious risks to user privacy, data integrity, and device functionality. Traditional malware detection methods are often insufficient due to the evolving sophistication of modern malware. This study investigates the use of deep learning, specifically Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), for enhancing Android malware detection accuracy and robustness. In this work, there is a comprehensive framework that utilizes both static and dynamic analysis to extract extensive features from Android applications. Static analysis examines the application's code, permissions, and metadata without executing the app, while dynamic analysis observes the app's behavior during runtime. These features are fed into CNN and RNN models, which are adept at handling spatial and sequential data, respectively, capturing complex patterns and temporal dependencies in malware behavior. Extensive experiments on a benchmark dataset of benign and malicious Android applications show that machine learning models, particularly CNNs and RNNs, significantly outperform traditional signature-based detection techniques. In this work, the best-performing model achieved a detection accuracy of over 95% with a low false-positive rate, demonstrating the potential of CNNs and RNNs to provide robust and scalable solutions for Android malware detection. This research highlights the importance of continuous feature engineering and model updating to adapt to evolving malware tactics and suggests that hybrid approaches combining static and dynamic analysis can offer more comprehensive defense mechanisms against sophisticated threats. Future work will focus on real-time detection capabilities and integrating these models into practical security solutions for end-users.