A detailed map database containing attribute information of individual buildings is highly valuable and expected to be utilized in various fields, including urban planning, energy management, and disaster preparedness. However, obtaining such detailed map databases are significant difficulty, because of privacy concerns and their high cost. To address this issue, this research aims to construct an open building-level spatial database as the goal. In this study, as a first step toward achieving this objective, we developed a method to classify buildings into detached houses and other types of buildings by utilizing Foundation Geospatial Data and information derived from open data. First, we assigned explanatory variables to each building in the foundation geospatial data for Nagaoka City, Niigata Prefecture, and created training data using PLATEAU data as the ground truth. Based on this dataset, we developed a machine learning model to classify each building as either detached or other types of buildings. Furthermore, we extrapolated the machine learning model to Sanjo City, Niigata Prefecture. We selected buildings in a way that aligns with the number of detached households reported in the national census at the subregion level and identified these as detached houses. Finally, validation of the extrapolated results showed that the mean absolute error (MAE) at the subregion level was approximately 9 buildings, demonstrating that the model successfully reproduced the spatial distribution of detached houses and other types of buildings.