VAE-GAN for Robust IoT Malware Detection and Classification in Intelligent Urban Environments: An Image Analysis Approach

The Internet of Things (IoT) has revolutionised technology in intelligent urban environments. Meanwhile, security and privacy risks have emerged, including the presence of various malware, resulting in detrimental consequences. Generative attack networks (GAN) can not only build superior representations for complex and multi-dimensional data but also maximise prediction performance due to their min-max optimisation manner. This paper proposes a GAN approach, utilising an autoencoder (AE) as the generator and a transfer learning for the discriminator, to identify various types of malware threats that exploit the IoT network using RGB images collected directly from malware samples. The generator is built for effective data reconstruction, with different AE structures and denoising manner; the discriminator utilises a pre-trained MobileNet for maximised performance. Two well-known image classification models, VGG19 and Xception, are used for performance comparison. The experiment proves that the Variational AE-GAN is highly implementable and scalable for the malware classification task, in both detection performance and generalizability.