Convolutional Variational Autoencoders and Resampling Techniques with Generative Adversarial Network for Enhancing Internet of Thing Security

Convolutional Variational Autoencoders and Resampling Techniques with Generative Adversarial Network for Enhancing Internet of Thing Security

연구 분야: Networking

학회: Pattern Recognition and Image Analysis

The Internet of Things is a pivotal constituent of the contemporary technological revolution and has experienced expeditious expansion in recent times. The proliferation of Internet of Things devices has led to enhanced convenience and automation. However, the extensive deployment of Internet of Things devices has also engendered concerns regarding data privacy and security. Among various detection and prevention methodologies, deep learning is emerging as a prominent trend. This paper ultilizes convolutional variational autoencoders and resampling techniques for network attacks detection. The proposed methodology employs a hybrid data resampling technique to tackle the issue of imbalanced classes, followed by the implementation of a convolutional variational autoencoder classification model with a weighted loss function. The experiments demonstrate that the light-weighted convolutional variational autoencoder outperforms the baseline models. Therefore, it possesses the capability to effectively detect intrusive activities in real-world settings and strengthen the Internet of Things security.