HOUSE PRICE PREDICTION USING MACHINE LEARNING WITH THE LINEAR REGRESSION ALGORITHM

Abstract

Data analysis enables accurate house price prediction by utilizing historical data and machine learning techniques. Determining house prices is essential for sellers, buyers, and property developers; however, it remains challenging due to the influence of various physical and socioeconomic factors. This study aims to predict house prices based on the variables CRIM (crime rate), LSTAT (percentage of the lower socioeconomic population), RAD (road accessibility), and RM (average number of rooms), with MEDV as the target variable. The Linear Regression algorithm was implemented using Orange Data Mining software, while the model performance was evaluated using the coefficient of determination (R²), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results indicate that the model achieved an R² value of 0.624, an MAE of 3.904, an RMSE of 5.574, and a MAPE of 0.202. Pearson correlation analysis revealed that RM has a strong positive correlation (+0.791) with house prices, whereas LSTAT has a strong negative correlation (−0.731). Meanwhile, CRIM (−0.389) and RAD (−0.384) exhibit moderate negative correlations with house prices. These findings suggest that both physical factors, such as the number of rooms, and socioeconomic factors, including crime rate and the proportion of the low-income population, significantly influence house prices. This study provides valuable insights for property developers and home buyers in considering these factors when determining property prices or making purchasing decisions. Future research is recommended to incorporate additional variables, more advanced machine learning algorithms, and larger datasets to improve prediction accuracy.