A Hybrid Consensus-Based Feature Selection Framework for Robust Housing Price Prediction

Abstract

Accurate predictions are essential for investment planning and risk assessment in real estate markets. Machine learning techniques have increasingly been adopted to improve the accuracy of housing price prediction. However, numerous studies have reported the absence of comprehensive predictor sets, ineffective feature selection practices, and limited model reliability and interpretability. Accordingly, this study develops a hybrid feature selection framework that compiles filter, wrapper, and embedded regularization feature selectors to improve the reliability of predictors. The framework employs the Boston Housing benchmark dataset and applies consensus-based feature selection prior to training four regression models: Linear Regression, Support Vector Regression, Random Forest, and Artificial Neural Networks. Four regression models were evaluated using the Mean Absolute Error (MAE), Mean Squared Error (MSE), the Root Mean Squared Error (RMSE), and the coefficient of determination to assess performance, reliability, and overall soundness of the models. Random Forest Regression achieved the highest predictive accuracy followed by ANNs (Artificial Neural Networks), SVR (Support Vector Regression), and standard Linear Regression. Robustness was achieved through the Random Forest Regression, which also exhibited the lowest error variance. As noted in the empirical evaluation, the models exhibit varying levels of performance. Random Forest Regression achieved maximum predictive performance and minimum error (MAE = 2.11, RMSE = 3.14) and surpassed the performance of the models of Artificial Neural Networks, Support Vector Regression, and Linear Regression. The findings support the assertion that consensus based feature prioritization and ensemble learning provide a consistent and reliable structure for modelling the housing price.