Large-scale modeling for housing condition prediction using machine learning algorithms

Area of Science:

• Environmental Science
• Urban Planning
• Data Science

Background:

• Housing price prediction is common, but large-scale housing condition prediction is limited by data availability.
• Existing research has not fully explored the spatial variations in housing quality across the United States.
• Understanding housing conditions is crucial for various societal applications.

Purpose of the Study:

• To develop and validate a machine-learning model for predicting housing conditions at a national scale.
• To overcome data limitations in assessing large-scale housing quality.
• To create a comprehensive dataset for spatial analysis of housing conditions.

Main Methods:

• Integrated property-level data (Warren Group) with U.S. Census Bureau neighborhood data.
• Trained and compared three gradient-boosting algorithms: CatBoost, LightGBM, and XGBoost.
• Selected CatBoost as the best model due to its superior resistance to overfitting.

Main Results:

• The CatBoost model demonstrated strong predictive performance for housing conditions.
• Predictions were aggregated to census tracts, ZIP code tabulation areas, and a hexagonal grid for spatial analysis.
• A comprehensive dataset for national-scale housing quality analysis was generated.

Conclusions:

• The developed machine-learning model effectively predicts national housing conditions.
• The resulting dataset is a valuable resource for analyzing the geography of housing quality.
• Applications include urban planning, disaster management, community resilience, and public health.