Abstract

This thesis builds on a growing body of research that seeks to understand how the built-in environmental attributes of the road network influence pedestrian route choice. Better understanding of these factors can help promotion of walkability. The thesis uses a high-quality GPS dataset of pedestrian trips recorded between October 17 to November 21, 2016, through the MTL Trajet app developed at Concordia University. Trip route characteristics are obtained by matching the GPS traces to a detailed GIS network dataset of road attributes. Additionally, built-in environment factors were captured by scenery quantification and micro-level land use analysis using Google Places API. Scenery was quantified by employing computer vision and machine learning techniques, with help of Google Street View API and deep learning frameworks. A path-size multinomial logit model is used to assess the utility of road and user features. Additionally, to improve prediction accuracy, a set of supervised learning classification techniques, including decision tree, random forest and gradient boosting tree were examined. The analysis of the results shows that the variation in scenery has a significant impact on pedestrians route choice. Additionally, machine learning classification techniques showed significant improvement of the accuracy ratio in comparison to discrete choice modeling framework.