Estimating the Spatial Distribution of Vacant Houses with Machine Learning using Municipal Data

In recent years, the number of vacant houses in Japan has continued to increase throughout the country, and understanding their distribution is an important problem for local governments. However, the method of surveying the distribution of vacant houses is primarily based on visual inspection from the outside, which requires much time, labor, and a budget for the survey. In this study, we developed a method to rapidly estimate the distribution of vacant houses in a municipality by developing a database of vacant houses in Maebashi City, Gunma Prefecture, a typical local city of Japan, by integrating a digital housing map and pinpoint data (basic resident register, water usage, and fixed asset taxation register) owned by the municipality, and performing machine learning using actual information on the distribution of vacant houses as ground truth data. We developed a method to rapidly estimate the distribution of vacant houses over the entire municipality.