Nonlinear relationship between microenvironmental exposure and travel satisfaction explored with machine learning

Abstract

Travel plays an indispensable role in daily life with regard to both individual health and subjective happiness. Using real-time environmental exposures and questionnaire survey data collected in Beijing, we apply the gradient boosting decision tree (GBDT) and Shapley value machine learning methods to investigate the nonlinear relationship between microenvironmental exposure and travel satisfaction. The results show that microenvironment-related indicators collectively contribute to over 50 % of the predictive power for travel satisfaction, while controlling for trip feature and socio-demographic variables. Real-time air and noise exposures during trips show threshold effects on travel satisfaction, with threshold values for air pollution and noise being estimated as 35 μg/m3 and 61db (A), respectively. This study also investigates the complex interactions that exist among pollution during travel, exposure at the residential location and attention to pollution, and in so doing explores the potential varied risks to different groups and provides policy implications for the government.