Analyzing the Nonlinear and Interaction Effects Mechanisms of Factors Influencing Intercity Travel Route Choice by Passenger Car Drivers: A Case Study from Guangxi, China

Although travel route choice analysis is abundantly conducted using statistical models, some shortcomings, such as easy neglect of nonlinear effects among factors, and subjective assumptions in modeling, are apparent. This study provides an interpretable framework based on machine learning models to better analyze the travel route decisions of intercity travelers. Four types of travel route choice behavior analysis models (i.e., extreme gradient boosting XGBoost, Light gradient boosting machine, random forest, and the binary logit model) were conducted using travel survey data for passenger car drivers in Guangxi, China, in 2021. The model parameters showed that XGBoost achieved the highest prediction accuracy (89.8%). Based on objective data distribution, the Shapley additive explanation approach was used to explain the output of XGBoost. The results showed that vehicle types, passenger capacity, expected toll discounts, and travel frequency on freeways had nonlinear effects on travel route choice, while traditional statistical models could not identify the nonlinear effects because of the effect of data distribution. Travel route choice was affected by potential interaction effects (e.g., vehicle types and toll payers). There were differences in the contribution of the same factor (e.g., education level) to route choice for different vehicle groups. These findings help better understand the generative mechanisms of travel route choice from a more objective perspective and provide references for developing more effective strategies to alleviate intercity road congestion and improve road network capacity by guiding travelers’ route choices.