Can Algorithms Predict Real Estate Prices? Multiple Regression Model Analyses

Resumo

The real estate sector plays a pivotal role in economic growth, making accurate property appraisal predictions essential for informed decision-making and investments. This study aimed to evaluate and compare the performance of supervised learning algorithms - Classification and Regression Trees (CART), K-Nearest Neighbors (KNN), Multiple Linear Regression (MLR), Support Vector Machine (SVM), Random Forest (RF) in predicting property values using a dataset from Recife, Brazil, spanning 1915 to 2024. Key property characteristics were selected using attribute selection, and models were assessed using R², MAE, MSE, and RMSE metrics. RF emerged as the most robust model, achieving a strong balance between accuracy and generalization, while SVM exhibited poor performance with large errors and limited predictive capability. Although MLR achieved the highest R², it struggled with inconsistent predictions. These results underscore the importance of algorithm choice and the influence of data characteristics, such as correlations and variable distributions, on model performance. This study contributes to real estate analytics by providing insights into effective machine learning applications for property value prediction, supporting both academic research and practical decision-making in the sector.