Predicting Resale Prices using Random Forests with Fine-Tuning Hyperparameters

Abstract

The accurate prediction of housing prices is essential for informed decision-making by purchasers, sellers, and policymakers in dynamic real estate markets. This study investigates the application of machine learning models—Random Forest, XGBoost, Decision Tree, and LightGBM—to predict resale flat prices in Singapore. It provides valuable insights into the use of machine learning in housing markets, particularly for datasets with similar size, complexity, and data types. The objectives are to develop predictive regression models for property prices and to analyze and compare the performance of these models. Key contributions include the development of tools to objectively estimate suitable property prices and the advancement of price prediction research through an extensive comparison of machine learning models. While previous studies have demonstrated the predictive capabilities of these models, this research focuses on the impact of hyperparameter tuning on the performance of the Random Forest model. By systematically optimizing parameters such as max_depth, n_estimators, and n_jobs, computation time was reduced by over 93% (from 865 seconds to 50 seconds) with minimal loss in accuracy. With proper hyperparameter tuning, Random Forest achieved the best performance in terms of MAE score (26.555), outperforming XGBoost (27.552), Decision Tree (28.832), and LightGBM (29.752).

Keywords

• Data Science
• Random Forest
• Hyperparameter Tuning
• Housing Price Prediction
• XGBoost
• Decision Tree