The present study proposes the novel use of integrated conformal prediction-based uncertainty estimation with machine learning algorithms for urban traffic noise prediction within a user-defined probability for traffic noise modelling. Three conformal prediction-based approaches — splitCP, CV+, and conformal quantile regression —were integrated with five machine learning models: GB, XGBoost, LightGBM, GPBoost, and NGBoost — to develop a statistically rigorous urban traffic noise level prediction in terms of equivalent noise levels (
). The dataset, consisting of 228 field measurements of
values with corresponding traffic flow per hour (
), was used. Of the total dataset, 168 samples were used for training, and the remaining 60 samples were used for testing. To determine the optimal value of user-defined parameters required by different machine learning algorithms, the AutoSampler package within Optuna, a Python-based hyperparameter optimisation framework, was used. To assess the quality of conformal prediction-based machine learning models, two criteria – mean predicted interval width and effective coverage were used. Results in terms of correlation coefficient, mean absolute error, and root mean square error values with the test dataset suggest improved performance by LightGBM and GPBoost using optimal values of user-defined parameters. A comparison of mean predicted interval width values suggests superior performance by the CV+ approach. Higher coverage values coupled with lower mean predicted interval width suggest that the CV+ conformal prediction approach is the best approach with the used dataset. Plot of predicted interval with best performing machine learning model at a miscoverage level of 0.1 suggests the effectiveness of the conformal prediction approach for uncertainty analysis in urban traffic noise prediction. Additionally, the practical applications of the study have been discussed, which can enable policymakers in urban noise mitigation planning and decision-making in several ways.
