TOD typology and station area vibrancy: An interpretable machine learning approach

Abstract

Transit-oriented development (TOD) has great potential to foster vibrant communities through improved access to activities around station areas. Several studies have investigated station area vibrancy and associations with TOD built environment (BE). However, few have considered the nonlinear impacts and varying relationships across station types. Taking Nanjing, China as a case study, we aim to 1) identify types of metro station areas (MSAs) with a “node-place-functionality” model and 2) unravel the nonlinear BE-vibrancy relationships and variations across MSA types. We find that five types best characterize the TOD built environment and present different levels of vibrancy indicated by the Baidu Heat Index. The GBDT (gradient boosting decision tree) models reveal transit accessibility, intersection density and commercial service density as the top three predictors of daytime and nighttime vibrancy, all presenting nonlinear and threshold effects. We also find the predicting power of BE features differs significantly across MSA types. The nuanced analyses provide context-specific planning guidance.