Comparison of machine learning regression methods for mass real estate valuation

Abstract

Efficient management of real estate requires an objective assessment of their values by using scientific approaches. Valuation is key for value-related applications such as purchase and sale, taxation, expropriation, and urban regeneration. Mass valuation reduces time and costs by evaluating multiple properties simultaneously. Leveraging statistical analysis and predictive capabilities of machine learning enhances accuracy and speed in real estate valuation. This study focuses on applying many regression models for mass valuation of residential properties in Melbourne, Australia, aiming to improve accuracy and efficiency for stakeholders. Evaluating various algorithms, including Linear Regression, Decision Trees, Random Forest, Bagging, AdaBoost, Gradient Boosting, and XGBoost, on Kaggle's open data, performance metrics are calculated. Notably, ensemble methods like Random Forest and XGBoost consistently outperformed others by capturing nonlinear relationships of determinants and predicting the value accurately. Finally, applying the Inverse Distance Weighting (IDW) interpolation method, a real estate value map is generated for the study area. This study aims to uncover machine learning's role and limitations in real estate valuation by comparing the performance of different ensemble learning methods. The findings highlight the significance of advanced regression models in improving valuation practices, supporting decision-making, and enhancing market efficiency.