UntitlPredicting High-Risk Vacancy Areas in Detached Housing: A Machine Learning Approached Item

This research addresses the growing issue of vacant houses, which has become an increasingly pressing concern in South Korea and other urban areas worldwide. Utilizing a machine learning-based methodology, we predict high-risk areas for vacancy occurrences in detached housing across Jinju-si, South Korea. Our approach utilizes a range of models, including Generalized Additive Models (GAM) and Random Forest, in combination with spatial autocorrelation, to enhance the accuracy of vacancy predictions. By analyzing demographic, geographic, and building-related variables, we identify key factors influencing vacancy occurrences, such as building age, land prices, and proximity to environmental pollutants. Additionally, we incorporate spatial dependencies through a spatially lagged variable, which significantly improves the model's predictive accuracy. Our findings reveal a concentration of high-risk vacancy areas in older urban regions, with predictions indicating a further spread to adjacent areas. This study provides critical insights for policymakers involved in urban regeneration, demonstrating the potential of advanced machine learning models to accurately predict and effectively manage urban vacancy dynamics.