Anomaly Detection in Network Security Using Unsupervised Graph Neural Network

This study presents an innovative methodology employing GNNs for network intrusion detection, utilizing the KDD Cup 1999 dataset, a well-established standard for assessing intrusion detection systems. The dataset encompasses a diverse array of simulated assaults and legitimate connections, consisting of 41 attributes for each connection record. We utilize stringent data preparation methods and develop a graph representation of the network data to enhance the GNN model. The experimental findings indicate that the suggested GNN surpasses conventional classifiers, with an accuracy of 93.7%, a precision of 92%, and an Fl-score of 91.5%. These metrics indicate substantial enhancements compared to KNN, SVM, CNN, and RNN models. The results indicate that GNNs proficiently capture intricate linkages within network data, improving the capacity to detect hostile activity. This research advances cybersecurity by offering a strong and effective solution for real-time intrusion detection, emphasizing the efficacy of GNNs in combating increasingly sophisticated network threats and enhancing digital security.