An Ensemble Approach to Route Choice

Abstract

Understanding automobile drivers' route preferences allows for the calculation of traffic flow on network segments and helps in assessing facility requirements, costs, and the impact of network modifications. Much previous research employs logit-based choice methods to model the route choices of individuals, but machine learning models are gaining increasing interest. However, all of these methods typically rely on a single `best' model for predictions, which may be sensitive to measurement errors in the training data. Moreover, predictions from discarded models might still provide insights into route choices. The ensemble approach combines outcomes from multiple individual models using various pattern recognition methods, assumptions, and/or data sets to deliver improved predictions. When configured correctly, ensembles offer greater prediction accuracy and better account for uncertainties. To examine the advantages of ensemble techniques, a hybrid method, which combines labelling and link-penalty approaches, is applied to a high resolution road network to prepare the choice set for route choice modelling. A data set from the I-35W Bridge Collapse study in 2008, and another from the 2011 Travel Behavior Inventory (TBI), both in Minneapolis - St. Paul (The Twin Cities) are each used to train a set of route choice models. These models are combined with ensemble techniques. The analyses consider travellers' socio-demographics and trip attributes. Based on the results, using the 10 best labels, the choice set captures most observed trajectories. A new similarity measure, which considers overlap, attribute similarity, and spatial similarity between routes, is proposed to evaluate models. We conclude that ensembles, when properly applied, perform better than base models. Heterogeneous ensembles using soft voting outperform both base models and other ensemble rules, with unscaled weights in soft voting proving more robust.