Vehicular traffic noise prediction using soft computing approach.

Area of Science:

• Environmental acoustics
• Computational intelligence
• Urban planning

Background:

• Vehicular traffic is a major source of urban noise pollution.
• Accurate prediction of traffic noise is crucial for urban environmental management.
• Existing models may not fully capture the complexity of traffic noise dynamics.

Purpose of the Study:

• To develop and compare soft computing models for predicting hourly equivalent continuous sound pressure levels (Leq) in urban areas.
• To evaluate the effectiveness of Generalized Linear Models, Decision Trees, Random Forests, and Neural Networks for traffic noise prediction.
• To identify the most accurate model for forecasting traffic noise based on traffic parameters.

Main Methods:

• Four soft computing methods were applied: Generalized Linear Model, Decision Trees, Random Forests, and Neural Networks.
• Models were trained using input variables: traffic volume per hour, percentage of heavy vehicles, and average vehicle speed.
• Model performance was assessed using coefficient of determination, mean square error, and accuracy, with 10-fold cross-validation for the Random Forest model.

Main Results:

• The Random Forest model exhibited superior performance compared to Generalized Linear Models, Decision Trees, and Neural Networks.
• 10-fold cross-validation confirmed the stability and reliability of the Random Forest model.
• A t-test indicated a good fit between the developed models and field data.

Conclusions:

• Soft computing approaches, particularly Random Forests, offer a robust method for vehicular traffic noise prediction.
• The developed models provide valuable tools for urban noise management and mitigation strategies.
• Traffic volume, heavy vehicle percentage, and average speed are significant predictors of hourly traffic noise levels.