Comparison of Analytical and Machine Learning Models in Traffic Noise Modeling and Predictions

Abstract

This paper illustrates the applications of analytical models and machine learning methods to predict the equivalent continuous sound pressure levels (L_Aeq) along with 10-percentile exceeded sound levels (L₁₀) generated due to road traffic noise based on rigorous noise monitoring conducted at more than 200 locations in Delhi-NCR. Using the measured data, regression, back-propagation neural network, and machine learning models were developed, validated, and tested. The work represents that the developed models are suitable for reliable and accurate predictions of hourly traffic noise levels. A comparative study reports that the machine learning-based model outperforms the classical analytical models. Multiple linear regression models and three machine learning techniques, namely decision trees, random forests, and neural networks, were utilized for developing models that predict the hourly equivalent continuous sound pressure level (L_Aeq1h) and 10-percentile exceeded sound pressure level (L₁₀). The developed predicted models have been ascertained to show an accuracy up to ± 3 dB(A). The proposed prediction models in the study can serve as a tool for planning noise abatement measures and traffic noise forecasts for the Delhi-NCR region. This study is the first rigorous study of its kind that covers a larger number of areas and zones in Delhi-NCR for assessment and predictions of road traffic noise and also shows an illustrative example of estimating measurement uncertainty in hourly noise measurements as per ISO 1996-2:2017.