Geographically Weighted Machine Learning for Modeling Spatial Heterogeneity in Off-Campus Student Housing Rents

Abstract

This study employs Geographically Weighted Machine Learning (GW-ML) to analyze spatial patterns and influencing factors of off-campus rental prices near colleges in Taiwan. Using Moran’s I, LISA, and K-means clustering, the research identifies rent clusters and segments the market. OLS, GWR, and GW-ML models were compared in terms of their performance, with Geographically Weighted LightGBM showing the best predictive accuracy. Key factors affecting rents include proximity to colleges, housing age, and access to public transportation. The GW-ML framework captures spatial heterogeneity and nonlinear relationships more effectively than traditional models. The results reveal that rental properties in city centres place more importance on rental area, while those in suburban areas focus more on building age for improving housing affordability and guiding policy in college neighborhoods.