Predicting Fine Spatial Scale Traffic Noise Using Mobile Measurements and Machine Learning.

Area of Science:

• Environmental Health
• Urban Planning
• Acoustics

Background:

• Environmental noise, particularly from traffic, is linked to adverse health outcomes like cardiovascular disease and sleep disturbances.
• Traditional noise monitoring methods struggle to capture the fine-scale spatial variations of traffic noise in urban environments.

Purpose of the Study:

• To develop and validate a machine learning model for predicting traffic-related noise with high spatial resolution.
• To assess the effectiveness of a novel on-foot mobile noise measurement technique combined with machine learning.

Main Methods:

• Collected A-weighted, equivalent noise (LAeq) data during hour-long foot journeys across 16 locations in Long Beach, California.
• Trained four machine learning models (linear regression, random forest, extreme gradient boosting, neural network) using traffic metrics, road features, meteorological data, and land use.
• Utilized leave-one-route-out and 5-fold cross-validation for model performance assessment.

Main Results:

• The extreme gradient boosting model demonstrated superior predictive accuracy, achieving an R-squared of 0.71 (RMSE 4.54 dB) in leave-one-route-out validation and 0.96 (RMSE 1.8 dB) in 5-fold validation.
• Local traffic volume was identified as the most significant predictor of noise levels.
• Road features, land use, and meteorological factors (humidity, temperature, wind speed) also contributed to noise prediction accuracy.

Conclusions:

• A novel mobile noise measurement strategy combined with machine learning enables precise prediction of small-scale spatial noise patterns.
• This methodology offers a powerful tool for urban environmental noise assessment and public health research.
• Accurate mapping of traffic noise can inform urban planning and mitigation strategies to reduce health risks.