Artificial intelligence models in real estate forecasting: a machine learning approach for sustainable urban construction in Saudi Arabia

Purpose The research aims to evaluate the predictive performance of three advanced machine learning (ML) models – extreme gradient boosting (XGBoost), ridge regression and categorical boosting (CatBoost) for real estate forecasting in Saudi Arabia, addressing the challenges of traditional valuation methods. Design/methodology/approach This study utilizes advanced ML models including XGBoost, ridge regression and CatBoost integrated with Z-score normalization and log transformation for preprocessing data from the Aqar website. Findings XGBoost demonstrated superior predictive accuracy with an R 2 of 0.98 (training), 0.77 (validation) and 0.82 (testing), outperforming CatBoost and ridge regression in terms of mean absolute error, mean squared error (MSE), root MSE and normalized root mean square error. Practical implications Stakeholders can benefit from enhanced transparency and interpretability in real estate decision-making, facilitating more informed and reliable investment strategies. Social implications Promoting sustainable development aligned with Saudi Arabia’s Vision 2030 through accurate forecasting supports long-term urban planning and economic stability. Originality/value This study uniquely evaluates the predictive performance of three ML models to improve forecasting accuracy in real estate, offering practical insights for sustainable urban planning and economic stability.