Developing vehicular traffic noise prediction model through ensemble machine learning algorithms with GIS

Abstract

<p>Vehicular traffic noise is an important aspect that requires environmental impact assessment (EIA) due to the increase in vehicular traffic. This has called for the need for detailed assessment, quantification and modelling of traffic noise pollution among industrial and scientific communities. This study uses an ensemble of machine learning algorithms to enhance the vehicular traffic noise prediction. The noise prediction is performed for the noise measure of the equivalent continuous noise level per 15 min (<i>L</i>_{eq, 15 min}) along the Subang Jaya and Shah Alam New Klang Valley Expressway (NKVE). In this study, three machine learning methods—(i) artificial neural network model (ANN), (ii) correlation-based feature selection with artificial neural network model (CFS-ANN) and (iii) ensemble machine learning algorithms random forest with artificial neural network model (Ensemble RF-ANN) are deployed to estimate the <i>L</i>_{eq, 15 min} during the peak hours of the morning (6:30–8:30 a.m.), evening (6:00–8:00 p.m.) and night (10:00 p.m.–12:00 midnight) for the weekday (Monday). The root mean square error (<i>RMSE</i>) and coefficient of determination (<i>R</i>²) were used to select the best model. The prediction maps for morning, evening and night are prepared using geospatial modelling for the two sites under consideration. The results showed that the Ensemble RF-ANN model is the best model recording the lowest <i>RMSE</i> during the training and testing with values of 1.767 and 2.378, respectively. It also achieved the highest <i>R</i>² values of 0.923 and 0.835, respectively, for the training and testing. The study provides novel noise models and excellent analysis of results capable of identifying key factors affecting noise in a geographical area. The study outcomes could be utilised by noise modelling consultants, town planners and traffic modelling experts for traffic noise mapping of larger geographical areas.</p>