Exploring passengers' choice of transfer city in air-to-rail intermodal travel using an interpretable ensemble machine learning approach

The transfer city is a key point in air-to-rail intermodal travel (ARIT) that directly influences the service level of the entire system. Although some studies have investigated factors that influence passengers' ARIT preferences based on subjective surveys, an in-depth understanding of their nonlinear and interactive impacts on passengers' actual behavior is still lacking. Using passengers' online booking data in China, this study implements an interpretable ensemble machine learning framework that incorporates decision-making theory to unveil feature importance and the complex nonlinear and interactive effects of various attributes on passengers' choice of transfer city in the crucial ARIT scenario. The results show that (1) the extreme gradient boosting (XGBoost) model achieves better performance in predicting ARIT transfer city choice than the conventional discrete choice model; (2) attributes related to intermodal services (e.g., ticket price, in-vehicle duration, transfer duration, quantities of flights and trains) are more important than personal demographic characteristics (e.g., age and gender); (3) factors related to service economy and efficiency display nonlinear impacts with fluctuations and critical thresholds; and (4) individuals with different characteristics present heterogeneous preferences for ARIT transfer cities. These findings can provide useful managerial implications for policymakers.

Automated summary

This study uses passengers' online booking data in China to analyze the factors influencing passengers' choices of transfer city in air-to-rail intermodal travel. The results show that intermodal service attributes are more important than personal demographic characteristics, and factors related to service economy and efficiency have nonlinear impacts. These findings are valuable for policy makers.