Pass-by Noise Modelling Applying Machine Learning

Overview & Synopsis

According to World Health Organization (WHO), traffic noise is, after air pollution, the second deadly environmental pollution in EU affecting human health. With the development of information and communication technology (ICT), more and more complex traffic data are generated and collected. Meanwhile the emerging field of data mining and machine learning brings data-driven methods under engineering attention. In this paper, standard pass-by noise data from Applus+ IDIADA is used for training and validating noise models that can predict the pass-by noise of a single vehicle. Different machine learning methods for regression, i.e. multiple linear regression, polynomial regression, decision trees and support vector machine, are applied to the IDIADA data for models training with K-Fold cross validation. Several models are trained separately based on different gears, driving conditions (wide open throat and cruise driving mode), and different masses of vehicles (lighter or heavier than 3500 Kg). A generalized model with integrating mass of vehicles as a categorical feature is also achieved. The accuracy (R2 , coefficient of determination) of noise prediction in the generalized model reaches up to 0.99 with applying support vector regression, decision trees and polynomial regression. These different algorithms are then compared with applying statistical tests, among which decision trees and SVR are proved to be not significant different and both are best models for the available dataset. In conclusion, this experiment of model training shows promising to apply machine learning in the field of transport, specifically in traffic noise study. Keywords: pass-by noise, machine learning, polynomial regression, multiple linear regression, decision trees, support vector machine, KFold cross validation, statistical test, null hypothesis, p-value.