Connected Queue Warning: Proactive traffic hazard warnings based on driving patterns

Abstract

This master’s thesis, conducted in collaboration with Volvo Cars, investigates how individual driving behavior can be used to predict routes and issue timely warnings about upcoming traffic congestion. The system is built using real-world trajectory data from the Geolife dataset, focused on Beijing. Routes are predicted using ant colony optimization , where pheromone weights are assigned to road segments based on previously traveled paths. Clustering is used to identify common start and end points. Destination prediction combines Random Forest, Bi-directional Long Short-Term Memory and a routing-based elimination method into an ensemble model that is continuously updated and maintains a strong prediction performance of above 69% to a maximum of 98% depending on the journey stage and the specific data split used for validation. The system sends simulated traffic jam signals which triggers warnings when the predicted path intersects with said traffic jams. To reduce false alerts, warnings are only sent when the model’s confidence is sufficiently high. The system, though tested on simulated traffic, provides a proof of concept and a foundation for future real-world applications.