A Study on Predicting Opportunity-Induced Boarding and Alighting Probabilities at Bus Stops Based on Spatial Context

Abstract

This study aims to predict opportunity-induced passenger alighting at bus stops by integrating spatial and statistical data in Ulsan, South Korea. Traditional models often miss spontaneous behaviors triggered by nearby POIs, such as shops or schools. Leveraging GIS and public datasets, the research defines a 500m influence zone around each stop to extract spatial features. A comprehensive data preprocessing pipeline merges location metadata, building geometry, and ridership statistics. Various regression models—including Random Forest, XGBoost and LightGBM—are applied and evaluated for predictive accuracy. Notably, even with only public data and basic preprocessing, the best- performing model achieved an explanatory power (R²) of approximately 0.64, suggesting strong potential as a baseline for transit demand prediction. Feature importance analysis highlights key POIs influencing demand, offering insights for data-driven transit planning.