Enhancing Housing Price Prediction Using AI and Machine Learning: A Stacked Regression Meta-Modeling Approach

Predicting housing prices is crucial in real estate markets, enabling informed decision-making for investors, buyers, and policymakers. Many studies have applied various machine learning models to this task, either focusing on individual models or overlooking the benefits of combining multiple models through ensemble techniques. Additionally, the neglect of optimizing model performance through feature selection and hyperparameter tuning often results in suboptimal outcomes. To address these gaps, we propose a meta-modeling and optimization workflow that integrates multiple regression models—such as Random Forest, Ridge, Lasso, and Support Vector Regression (SVR)—using a stacked regressor to improve prediction accuracy. Our workflow includes systematic hyperparameter tuning and feature selection to refine model performance. The stacked model outperforms individual models, achieving an R² score of 0.9321 and an RMSE of 1.5017, representing a 6.5% improvement in R² and a 39.2% reduction in RMSE compared to the best individual model, Random Forest. This demonstrates the effectiveness of ensemble methods in capturing complex relationships in housing data. Furthermore, our proposed workflow is flexible, allowing adaptation to various datasets and market conditions. This research highlights the importance of advanced model optimization and ensemble learning in housing price prediction, providing a robust solution that significantly improves accuracy over traditional models.