Streaming-Based Anomaly Detection in ITS Messages

Abstract

Intelligent transportation systems (ITS) enhance safety, comfort, transport efficiency, and environmental conservation by allowing vehicles to communicate wirelessly with other vehicles and road infrastructure. Cooperative awareness messages (CAMs) contain information about vehicles status, which can reveal road anomalies. Knowing the location, time, and frequency of these anomalies is valuable to road users and road authorities, and timely detection is critical for emergency response teams, resulting in improved efficiency in rescue operations. An enhanced locally selective combination in parallel outlier ensembles (ELSCP) technique is proposed for data stream anomaly detection. A data-driven approach is considered with the objective of detecting anomalies on the fly from CAMs using unsupervised detection approaches. Based on the experiments carried out, we note that ELSCP outperforms other techniques, with 3.64 % and 9.83 % better performance than the second-best technique, LSCP, on AUC-ROC and AUCPR, respectively. Based on our findings, ELSCP can effectively detect anomalies in CAMs.