Predicting Traffic Patterns Using V2X Communication Data

The aim of this study is to investigate whether vehicle to vehicle (V2V) communication data can be used to understand and forecast traffic patterns. The V2AIX V2X dataset contains over 230,000 messages collected from over 1,800 vehicles and roadside units in public road traffic, extracting instantaneous features, such as speed, bearing, and acceleration. Two unsupervised machine learning techniques of K-Means and Agglomerative Clustering were used to uncover hidden traffic patterns. They also construct methods that identify distinct clusters of vehicles travelling on particular routes in the same or opposite directions, travelling through intersections or curved road segments, and travelling in alternate or parallel road segments. The results suggest that clustering V2V data could improve traffic prediction models and lead to useful traffic management and optimization strategies.