Citation

Liu J, Xiao L. Transportation (Amst) 2024; 51(1): 1-24.

Copyright

(Copyright © 2024, Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s11116-022-10317-5

PMID

unavailable

Abstract

Hundreds of millions of internal migrants are present in cities in the developing world. Accurately predicting their commuting mode choice and clarifying the effect size of the influencing factors have become increasingly indispensable for formulating and executing socially inclusive transportation plans and policies. Yet, scholarship on internal migrants' travel mode choice is still scarce, particularly in the context of developing country. This study attempts to partly fill this gap. Using empirical data from Xiamen, China, it applies the Light Gradient Boosting (LightGBM) approach, a high-efficiency and high-performance machine learning framework, to predict the commuting mode choice of both the internal migrants and locals. The results show that (1) The built environment has larger impacts on locals' mode choice than on migrants'; (2) For both the migrants and locals, the relative importance of the built environment in predicting commuting mode choice exceeds that of socio-demographics and trip characteristics; (3) Distance to the closest commercial center is the most important factor influencing commuting mode choice of both groups, and bus stop density also contributes a great deal; hence, regional accessibility and transit infrastructure can be given higher priority in intervening the commuting behaviors of migrants and locals. (4) The LightGBM models yield rather high prediction accuracy; their results are further compared with those of conventional discrete choice models and are found to be generally consistent with the latter. Those findings can help inform decision-makers about nuanced policies concerning meeting locals' and migrants' travel demands.

Language: en

Keywords

Built environment; China; Commuting mode choice; Discrete choice model; Light gradient boosting; Migrants and locals