Android Malware Detection Using Generative Adversarial Networks: A Deep Learning Approach

Abstract

The Android operating system, the most widely used mobile platform in recent years, is a prime target for malware attacks. Conventional detection techniques have significant challenges due to the growing volume and complexity of Android malware, especially signature-based techniques that struggle with obfuscated code and zero-day threats. This article proposes a novel approach to Android malware detection that integrates Generative Adversarial Networks (GANs) with deep learning (DL) and machine learning (ML) approaches. The proposed methodology tackles two key problems: the lack of malware samples and the data discrepancy between malicious and benign datasets. By employing GANs for data augmentation, the work increases the number and diversity of training data, hence improving detection accuracy. Custom datasets were created by extracting static and dynamic properties from Android application packages (APKs) using tools like CuckooDroid. A Random Forest approach was employed to choose a total of 357 characteristics in order to ensure optimal model performance. The F1-score, accuracy, precision, and recall were used to evaluate the classification models that were trained on the GAN-generated data. Experimental results showed that the GAN-based model outperformed state-of-the-art methods with a high accuracy of 99.10%. These findings imply that GANs can greatly enhance Android malware detection systems by enhancing the caliber and variety of datasets.