Travel Time Prediction: Comparison of Machine Learning Algorithms in a Case Study

Travel time prediction has important applications within the field of intelligent transportation, such as vehicle routing, congestion and traffic management. A challenging task in travel time prediction is obtaining data that is not readily available, as a clear majority of links in roads network are not equipped with traffic sensors. In this paper, data of travel time is collected for a link using Google Maps Application Programming Interface (API). Then, travel times are predicted for short horizons of up to one hour on the link by applying machine learning algorithms. The Mean Absolute Error (MAE) of predictions are compared. The study indicates that a shallow Artificial Neural Network (ANN) can provide more accurate prediction than the other algorithms.

I. Introduction

Travel time information is nontrivial for travelers, transportation agencies and traffic management authorities. Information on the future state of traffic enables travelers and transportation agencies to plan their trips and avoid congested roads, leading to reduce transport costs and delays. Traffic management authorities using travel time information can apply policies, such as rerouting traffic or optimizing the signaling time of traffic lights, to reduce or prevent congestion. These measures also reduce greenhouse gas emissions, as CO₂ emission rates in congested conditions can be up to 40% higher than the emission rates seen in free-flow conditions [1].