Prediction of the median value of owner-occupied homes (MEDV) in the Boston housing dataset using regression models

The Boston housing dataset is one of the significant tools used to examine the influencing factors of housing price. Meanwhile, housing price prediction is crucial for government regulation, business decision-making, and individual homebuyers. Existing studies fall short in balancing model interpretability, computational efficiency, and generalization ability. Hence, this study, based on the Boston housing dataset, focuses on the prediction of the median value of owner-occupied homes (MEDV). It constructs three models, including linear regression, decision tree regression, and Bayesian regression, and evaluates their performance using mean squared error (MSE), mean absolute error (MAE), and the coefficient of determination (R2). It examines the effects 13 features, including per capita crime rate by town (CRIM) and average number of rooms per dwelling (RM) on MEDV. Furthermore, it trains and visualizes the results of each model. The results show that decision tree regression achieves the highest R2, effectively capturing nonlinear relationships but being prone to overfitting. Linear regression and Bayesian regression perform better in terms of MSE and MAE; the former is simple in structure and fast to train, while the latter can output probability distributions to assess uncertainty. Each model has its strengths, and the choice should depend on the application scenario. The limitations related to dataset timeliness and the lack of extensive hyperparameter tuning are acknowledged, providing useful insights for housing price prediction research and practice.