A Method for Estimating Building-Level Vacancy Probability Based on Exterior Images and Building / Neighborhood Attribute Data

This paper developed a method to estimate the probability that each building is vacant at the building level, using diverse information, including building attributes and regional data, as features in addition to building exterior images. From building exterior images, we applied computer vision techniques to extract features such as component proportions, overall impression, structure, and construction year. Ground-truth data on vacant buildings were obtained through field surveys and resident questionnaires. Logistic regression analysis showed that our proposed method achieved an F1-score of 83.7% in identifying vacant buildings, demonstrating performance comparable to that of a comprehensive building survey.