An intelligent hybridized computing techniques for the prediction of roadway traffic noise based on non-linear mutual information

Abstract A reliable traffic noise prediction model is one of the decision-making tools used in providing a noise friendly environment. In this study, four linear-nonlinear hybrid models were proposed to capture both linear and nonlinear patterns of the data by summing up the predicted traffic noise from the multilinear regression (MLR) and estimated residuals from four artificial intelligence (AI)-based models. The input variables for the models were volumes of cars, medium vehicles, buses, heavy vehicles, and average speed. Prior to the development of the hybrid model, the potential of Boosted Regression Tree (BRT), Feed Forward Neural Network (FFNN), Gaussian Process Regression (GPR), Support Vector Regression (SVR) and Linear regression models for traffic noise prediction were evaluated and compared with each other. The performances of the single and hybrid models were evaluated using the Nash-Sutcliffe efficiency (NSE), root mean square error (RMSE), mean absolute error (MAE) and relative root mean square error (rRMSE). The results showed that, the hybrid models provide better prediction capability than both the linear and nonlinear models in both calibration and verification stages. MLR-GPR hybrid demonstrated better prediction skill than all other hybrid models with NSE, RMSE, MAE and rRMSE values of 0.9312, 0.0427, 0.0347 and 7.4%, respectively. The study found that, the efficiency of the linear models could be improved up to 27.26% when they are hybridized with the nonlinear models.