Ma, Yuxun (Tokyo Institute of Technology), Seo, Toru (Tokyo Institute of Technology)

Incorporating Graph Neural Network into Route Choice Model in Road Network

Scheduled for presentation during the Poster Session "Travel Behavior Under ITS" (WeAT16), Wednesday, September 25, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Route choice model is important for Intelligent Transportation Systems. The most commonly used route choice models include seminal logit models and their extension recursive logit model. The advantage of these models is their interpretable parameters and thus useful for policy making. Recently, more accurate models have been proposed by using data-driven approaches such as deep neural networks, however, they generally lacks interpretabllity. In this study, we proposed a hybrid model of recursive logit model and Directed Graph Neural Network which has the advantage of interpletable parameters as well as high accuracy of capturing complicated data in road network. Applying the model to actual travel trajectory data in Tokyo, our proposed model shows higher prediction accuracy compared to recursive logit model and a Residual Neural Network-based Recursive logit model. Moreover, the prediction accuracy and the interpretability can be balanced arbitrarily by adjusting the penalty coefficient. Finally, we proposed a LRP-based link importance analysis method to interpret the nonlinear neural network and detect which links affect the prediction accuracy most.