Evaluating the predictive power of ensemble, regularization and neural network models in property price forecasting in Prishtina

What the paper says

Purpose The accurate prediction of property prices remains a significant challenge in real estate market analysis. This study aims to compare the predictive performance of various machine learning (ML) models, including Random Forest, Gradient Boosting, Lasso Regression, Elastic Net, AdaBoost Regressor, Bayesian Regression, Bagging Regressor and Stacking Regressor. Design/methodology/approach Using a comprehensive data set of 1,512 property transactions from Prishtina, Kosovo, spanning 2019–2023, this study applies a consistent methodological framework for evaluating each model’s performance using Mean Squared Error, Coefficient of Determination (R²), Mean Absolute Error and Root Mean Squared Error. The models were trained on 70% of the data, validated through k-fold cross-validation and tested on the remaining 30%. Findings The results indicate that ensemble methods, particularly Random Forest and Bagging Regressor, exhibit superior predictive power with low error metrics and high R² values. While complex models show high accuracy, simpler linear-based approaches like Lasso and Elastic Net provide benefits in interpretability. The findings highlight the need to balance predictive performance with interpretability and computational efficiency. Originality/value This study’s insights support real estate practitioners seeking data-driven valuation methods and guide researchers aiming to enhance property price prediction in emerging markets. By offering a robust evaluation of diverse ML techniques in this context, the research contributes valuable knowledge for model selection and performance optimization.