Temporal variations in the non-linear relationships between metro ridership and the built environment: insights from interpretable machine learning using four-year data

Abstract

Listen

Translate article

Abstract Understanding the association between metro ridership and the built environment is crucial for promoting integrated transportation and land use planning. However, prior research has rarely examined the temporally varying and/or non-linear associations between metro ridership and the built environment. To address this gap, this study collects metro ridership data in Chengdu, China, for January of each year between 2019 and 2022 and uses light gradient-boosting machine (LightGBM) and SHapley Additive exPlanations (SHAP) models to examine the complex, non-linear associations between metro ridership and the built environment over four years. Our findings highlight the non-linear nature of the built environment’s influence. The key predictors remained relatively stable throughout the years, including the number of entrances (the top predictor across all years), employment density, and the floor area ratio. However, the influence of built environment factors, such as land-use mix, residential micro-district density, and distance to the city center, shows great temporal variations, underscoring the importance of incorporating temporal dynamics into analyses of the interactions between metro ridership and the built environment. This study offers a valuable reference for urban and transportation planners in crafting tailored policies for station-area transit-oriented development (TOD).