What type of study is this?

September 5, 2025Open Access

Modeling Real Estate Transaction Prices in Tokyo: The Value of Comprehensive Metadata in Unsupervised Clustering and Predictive Modeling Analysis

Key Points

The Random Forest model produced a best performance with an RMSE of approximately 185 million, indicating effective predictive modeling.
Key predictors such as unit price, floor area ratio, and building year significantly influence transaction prices in Tokyo's real estate market.
Inclusion of comprehensive metadata leads to improved accuracy and interpretability in urban real estate pricing models.
Limited datasets provide lower predictive power, suggesting structural features are vital for accurate real estate price predictions.

Abstract

Japanese modern real estate market has experienced drastic fluctuation since the explosion of the bubble economy. After few years of depression, the transaction price eventually converged to a stable state due to the stable household savings of business corporations and developed typical trends that were suitable for modeling. This study explores the determinants of real estate transaction prices in Tokyo, leveraging a large-scale dataset comprising over 400,000 observations. Among the various models evaluated, the Random Forest model had the best performance, with an RMSE of approximately 185 million and an R of 0.577 on the detailed set. In contrast, the limited dataset shows lower predictive power, with higher RMSE and lower R, highlighting the importance of structural features in price prediction. Key variables such as unit price, floor area ratio, land breadth, and building year consistently emerge as significant predictors. This research demonstrates that the inclusion of detailed building characteristics substantially improves model accuracy and interpretability in urban real estate modeling.

Ask AI

Helpful

Bookmark

View Full Paper