Predicting Housing Prices in Sweden: A Comparative Study of Linear Regression and Machine Learning Models

This study investigates the factors influencing the asking prices in the Swedish housing market, focusing on property-specific variables such as the number of rooms, land area, living area, and price per square meter. Using a linear regression model, the analysis reveals that these factors explain 18.9% of the variation in asking prices, with all coefficients showing significant positive relationships. To address the limitations of linear models, this study further employs nonlinear machine learning approaches, including decision trees and random forests, to capture complex interactions in the data. The decision tree model achieves perfect fit on training data (R² = 1.000) but shows reduced generalization on test data (R² = 0.800), suggesting potential overfitting. In contrast, the random forest model demonstrates robust performance, with high explanatory power (R² = 0.892 on test data) and minimal prediction errors, highlighting its superiority for housing price forecasting. Diagnostic tests confirm the absence of multicollinearity and autocorrelation in the linear model, while the machine learning models provide deeper insights into nonlinear relationships. These findings offer valuable guidance for investors and policymakers, emphasizing the importance of model selection in housing market analysis. Future research could integrate spatial analytics and macroeconomic shocks to further improve predictive accuracy.