GAN-based Enhancer for Malware Detection on Android Platform

Abstract

Mobile malware applications have become a prominent approach for attackers to target vulnerable users since smartphones have become widespread among people during the last decade. Recent research efforts to detect and classify malware have applied Artificial Intelligence (AI) approaches. However, the difficulty in collecting mobile malware samples and their tendency to show an imbalance between malware families insert a significant roadblock in utilizing AI techniques, particularly Deep Learning (DL) methods, for mobile malware detection. This thesis proposes using Generative Adversarial Networks (GAN) to counter these challenges and enhance the performance of Machine Learning (ML) and DL-based detection models. Several experiments were carried out to determine the most suitable multiplication factor of the original data generated by the GAN-based enhancer, and two Conditional Tabular GAN-based synthesizers were used and compared to identify the one with better results. The obtained results demonstrate the ability of the applied data augmentation to improve the performance of the mobile malware detection models, particularly the ones based on DL techniques. The results also indicate that doubling and tripling the original samples was sufficient to attain the best results for the ML and DL models, respectively. The CTGAN synthesizer produced higher-quality samples of the two compared synthesizers, leading to a stable performance compared with the CopulaGAN synthesizer.

Date of Award	9 Dec 2024
Original language	American English
Supervisor	Chan Yeun (Supervisor)