Traffic Delay Prediction From Historical ARCHES

Road traffic congestion is one of the biggest frustrations for the daily commuter. By improving the currently available travel estimates, one can hope to save time, fuel and the environment by avoiding traffic jams. Before one can predict the best route for a user to take, one must first be able to accurately predict future travel times. In this thesis, we develop a classification-based technique to extract information from historical traffic data to help improve delay estimates for road segments. Our techniques are able to reduce the traffic delay prediction error rate from over 20% to less than 10%. We were hence able to show that by using historical information, one can drastically increase the accuracy of traffic delay prediction. The algorithm is designed to enable delay prediction on a per-segment basis in order to enable the use of simple routing schemes to solve the bigger problem of predicting best future travel paths. Acknowledgments I would like to thank Arvind Thiagarajan for his guidance and insight in parts of this project. I would also like to thank him for providing the map-matched data of taxicab trajectories that is used for travel delay estimation.