Project on Anomaly Detection using Graph Neural Networks

The project involved building a Graph Neural Network model that identifies sensor values on heterogeneous graph data consisting of different types of entities and then makes a link prediction to detect any anomaly patterns that deviate from regular observed patterns over the Graph Convolutional Network based on a time series activity trend and network architecture. The graph data are structurally complex as each node in the network represents relationship edges between nodes of different types of entities linked in the graph.

To achieve the expected results, I utilized machine learning to solve the complex task by performing node classification, link prediction and graph classification.

The project procedure involved leveraging GNN-based methods which learn the presented graph attributes, including structure and features then make appropriate assignments in scoring anomalies.