In Japan, the rise in vacant housing has created the need to develop quick, effective, and inexpensive methods to detect the spatial distribution of vacant housing at the municipal level. However, due to incomplete and inaccessible data, the change in the accuracy of the vacant housing detection model must be evaluated while accounting for the limited data. Therefore, this study compares the performance of vacant housing detection models for different data combinations (Basic Resident Register; building registration, water usage, and national census) by considering Wakayama City, Japan, as the case study setting. Three main findings emerged: (1) the contribution of the data to the accuracy varies with the combination of datasets and metrics; (2) even if specific municipal data are unavailable, it is possible to acquire a similar accuracy by combining other data; and (3) the missing value contributes to the vacant housing detection rather than the feature value itself.

No abstract