Predicting highly dynamic traffic noise using rotating mobile monitoring and machine learning method

Area of Science:

• Environmental Science
• Acoustics
• Urban Planning

Background:

• Traffic noise is a significant global environmental issue, posing challenges for urban management.
• Generating dynamic traffic noise maps is hindered by data scarcity and predictive limitations.

Purpose of the Study:

• To introduce an innovative noise monitoring method for improved data acquisition.
• To develop accurate predictive models for traffic noise levels.
• To create high-resolution, dynamic traffic noise maps.

Main Methods:

• A novel Rotating Mobile Monitoring approach was implemented, combining stationary and mobile data collection.
• Extensive data, including A-weighted equivalent noise (LAeq), street view images, and meteorological data, were gathered.
• Machine learning models, including Random Forest and K-Nearest Neighbors, were trained using 49 predictor variables.

Main Results:

• The Random Forest model achieved the highest predictive accuracy (R² = 0.72, RMSE = 3.28 dB).
• Key predictors for noise levels included distance to major roads, tree view index, and car field of view.
• A 9-day dynamic traffic noise map was successfully generated for the study area.

Conclusions:

• The Rotating Mobile Monitoring method effectively expands spatial and temporal noise data coverage.
• Machine learning models provide a reliable approach for predicting traffic noise.
• The methodology is scalable and replicable for creating dynamic noise maps in diverse urban environments.