Anomaly Detection in In-Vehicle Networks with Graph Neural Networks

In today's vehicles, various functions required for the vehicle are managed by Electronic Control Units (ECU) and the communication of these units takes place over the in-vehicle network. With the increase in the number and complexity of vehicle functions, the number of ECUs also increases and in-vehicle network message traffic becomes more complex. Detection of anomalies in in-vehicle network message traffic for the detection and prediction of problems in vehicles has become an important research problem. In the literature, time series analysis based solutions are suggested for this problem. On the other hand, graph-based machine learning and anomaly detection studies have come to the fore recently. In this study, a graph neural network (GNN)-based solution is applied for anomaly detection on in-vehicle network messages. The analyzes on the driving simulation data showed that the GNN-based solution produces successful results for anomaly detection on in-vehicle networks.