Adaptive Anomaly Detection in Dynamic Graph Networks

Over the past few years, the detection of anomalies in dynamic graph networks has attracted substantial attention worldwide because of its applications in various fields such as cybersecurity, financial fraud detection, and social network analysis. Traditional methods may not be as adaptive to the dynamic nature of such networks, as it requires more efficient and scalable solutions. To mitigate the issue, a Variational Graph Convolutional Network (V-GCN) is proposed. It is an approach that combines variational autoencoders (VAEs) with graph convolutional networks (GCNs) for anomaly detection in dynamic graph data to enhance anomaly detection. The methodology involves developing a V-GCN architecture that combines the capabilities of VAEs and GCNs, applying an adaptive learning strategy to update model parameters in real time, and carrying out experiments using standard datasets. The experimental results indicate that the anomaly detection performance considerably enhanced using the proposed model. It outperforms the baseline methods, where the V-GCN attains an AUC-ROC of 95%, an average precision of 92%, an F1 score of 91%, and an anomaly score of 94%.