Comparison of Selected Regression Models in Predicting Railway Traffic Noise Levels

The aim of this study was to compare the effectiveness of selected regression methods in predicting the noise level generated by railway traffic. The analysis was based on measurement data collected at ten locations in Poland, taking into account technical and environmental parameters as well as train passage characteristics. Linear regression (OLS), Weighted Least Squares (WLS), LASSO, Ridge and Elastic Net were used in the modelling, and their effectiveness was assessed using the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). The results showed that the noise level depends primarily on train speed, passage time, and rolling stock type, while meteorological variables had a marginal impact. The best fit was obtained for the WLS model, with effectively solving the problem of heteroscedasticity. Regularized models made it possible to reduce the number of predictors without losing the quality of the fit. The study confirms that modern regression techniques can be a valuable tool for assessing the impact of railways on the acoustic environment.