Anomaly Detection in Graph-Based Data Utilizing Graph Topology

This paper presents a novel framework for anomaly detection in complex networks using Graph Neural Networks (GNNs). By incorporating graph topology and node attributes, the proposed model offers improved accuracy and detection capabilities compared to traditional methods. The proposed framework utilizes a graph autoencoder with decoders for node features and the adjacency matrix, capturing the underlying structure of the graph. Topological features such as centrality and clustering coefficients are integrated to enhance the model's performance. To address the complexities of the model, a generic algorithm is employed for efficient optimization. Our approach demonstrates superior performance compared to traditional autoencoders and other machine learning models. The model's effectiveness in identifying anomalous nodes and subgraphs highlights the importance of considering both structural components and node attributes in anomaly detection with GNNs. One of the key advantages of the proposed model is its adaptability to various datasets and node features. Training the model with multiple graphs or instances allows for handling diverse working conditions and uncertainties.