Citation

Dong G, Kweon Y, Park BB, Boukhechba M. J. Big Data Anal. Transp. 2022; 4(2): 119-133.

Copyright

(Copyright © 2022, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s42421-022-00058-3

PMID

unavailable

Abstract

GPS-based navigation systems have played crucial roles to improve transportation system performances. A limitation of such route guidance systems is that their route recommendation algorithms are mostly based on the shortest travel time and/or toll avoidance. To enhance the performance of route recommendation systems, it is critical to model drivers' route choice behaviors, such that personalized route recommendations can be delivered to individual drivers or groups of drivers who share similar route choice behaviors. Previous studies have focused on game theory-based discrete choice models (e.g., Multinomial Logit) or shallow machine learning models (e.g., support vector machine) to generate the interpretations of route choice behavior by using the estimated parameters. Despite the good interpretability of these models, such interpretation could be often biased due to a small number of observations, restricted model complexity, and incapacity to capture the rich variations in route choice behaviors. In this paper, by leveraging both route level and route segment level features, we apply deep sequential models for the route data , and propose a hybrid sequence-to-one deep neural network model, named hybrid transformer-LSTM (HTL), to model the complex route choice behaviors. We utilize a large real-world route choice dataset to evaluate HTL and demonstrate its performance as compared to other state-of-the-art sequential neural network models. A model agnostic interpretability method, called SHAP, is applied to explain the route choice rationals. Finally, a benchmark clustering of route choice behavior is demonstrated to unveil the latent heterogeneous partitions in route selection decision making, based on the shapely values provided by SHAP.

Language: en

Keywords

Decision making; Perception; Route choice behavior; Transformer; Transportation; Utility theory