Estimation of Travel Time Index Using Machine Learning Methods

Travel time and its variation are among important aspects of transportation that are used as key indicators to evaluate network performance in transportation planning. Studies show that traffic congestion is an important factor effecting travel time reliability and is divided into two categories: Recurrent and Non-Recurrent. Studies show that Traffic incidents, Work zones, Weather, Demand fluctuations, Special events, Traffic control devices and bottlenecks are the seven major factors which cause travel time changes. This article aims to examine how changes in transit geometry (number of lanes), accidents, traffic volume, and weather conditions can affect travel time reliability. In order to do so, a variety of machine learning methods were used to study and model the Virginia highway network, including Support Vector Regression, Nearest Neighbor regression, and Decision Tree regression. The results of this study showed that these tools can reflect the changes in the average travel time to an appropriate extent. Among these methods, performance of the Nearest Neighbor was the best (coefficient of determination 0.85 and error 0.0012 for training set and coefficient of determination 0.57 and error 0.0014 for test set and stable performance equal to 1.47).