A Lightweight Generative Adversarial Network for Imbalanced Malware Image Classification

Classifying malware images is important in cybersecurity. The nature of different families and classes of malware images is imbalanced, which leads to a challenging issue of biased classification, where the majority classes dominate the model's performance. Attention is drawn to generative artificial intelligence, in which a generative adversarial network (GAN) is used to synthesize more training samples (particularly in minority classes). A more balanced dataset can reduce biased classification towards the majority class. However, a robust GAN model usually requires high computing power and a lot of epochs. In this paper, we aim to design a lightweight GAN model to enhance the performance of malware image classification in imbalanced datasets. Performance evaluation and analysis show that it enhances the model's accuracy and reduces its training time. Future research directions are also shared.