Predicting Real Estate Prices with AI: A Comparative Study of Machine Learning Models

Abstract

Accurate house price prediction is vital for economic, financial, and policy decision-making, impacting homebuyers, investors, financial institutions, and government agencies. This study employs a data-driven machine learning approach to forecast residential property prices, with a particular focus on high-rise properties in Kuala Lumpur. Real-world housing data based on 12,735 transactions from 2021 to August 2024, were collected from the National Property Information Centre (NAPIC), pre-processed, and analysed using Exploratory Data Analysis (EDA) to understand the influence of various property attributes on prices. Multiple predictive models, including traditional regression, ensemble methods (Random Forest, Gradient Boosting Machine), and deep learning (Artificial Neural Networks), were developed and rigorously compared. Model performance was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R²) on an 80:20 training-testing split. Hyperparameter tuning and K-fold cross-validation were applied to optimize accuracy, prevent overfitting, and ensure model generalizability. The Random Forest model emerged as the optimal predictor, demonstrating robust performance with the lowest error values and highest R² score compared to other tested algorithms. This research provides practical insights into key price-influencing features and highlights the efficacy of machine learning for robust and interpretable house price prediction in the Malaysian real estate market.