The valuation of residential properties is key for effective real estate market operation, mortgage lending, and investment decisions. Multiple Linear Regression (MLR), along with other traditional approaches, was widely employed to calculate the values of real estate because of its transparency and ability to understand the process behind predictions. However, traditional models face problems with capturing non-linear relations which characterize modern housing markets. Machine Learning (ML) approaches, including such algorithms as RF and XGBoost, show better results in terms of predictive accuracy, yet they lack interpretability. This paper aims at comparing predictive power and interpretability of MLR, RF, and XGBoost based on the dataset of Texas Residential Real Estate Intelligence 2026 comprising 12,137 entries about residential properties for sale. The methodology of the research will involve quantitative analysis. Data will be collected using secondary sources specifically, Kaggle. Performance of ML models will be measured by means of R², RMSE, and MAE. To evaluate interpretability of models, SHAP will be utilized. In addition, the study aims to make a contribution to the existing literature on explainable AI by investigating if SHAP gives more insight into prediction than regression coefficients alone. Moreover, the study shall give an indication of the trade-off that exists between predictive accuracy and explainability in home value prediction models.
Thato Mashiane (Fri,) studied this question.