Network Intrusion Detection Optimization based on Graph Neural Networks and Variational Autoencoders

In recent years, there has been a notable increase in the prevalence, diversity and complexity of cyber attacks. Traditional detection methods have proven to be inadequate in effectively capturing new attack traffic, resulting in missed detection and economic losses. It is therefore of particular importance to enhance the efficacy of the network intrusion detection system in order to discern potential attack behaviours by identifying anomalous traffic patterns. This paper puts forth an innovative optimization model for network intrusion detection based on deep learning. The model proposes a novel deep neural network architecture that integrates the strengths of graph neural networks (GNNs) and variational autoencoders (VAEs). Firstly, a graph structure representation of network traffic is constructed using GNN in order to capture the complex relationships and potential patterns between nodes. Subsequently, a variational autoencoder (VAE) is employed to extract and reduce the high-dimensional features of the graph structure data, thereby retaining the key features while reducing the influence of noise. Furthermore, an adaptive attention mechanism is incorporated to facilitate the dynamic assignment of feature weights, thereby enhancing the sensitivity of the model to anomalous behaviours. The efficacy of the proposed model was evaluated through experimentation using the UNSW-NB15 and CIC-IDS datasets. The results demonstrated that the model exhibited superior performance in terms of detection accuracy and recall compared to existing methods.