An automated framework for traffic noise level analysis using explainable artificial intelligence techniques.

Area of Science:

• Environmental Science
• Artificial Intelligence
• Urban Planning

Background:

• Traffic noise is a major urban pollutant, with current prediction methods often lacking interpretability.
• Advanced machine learning (ML) models offer potential for more accurate traffic noise prediction.

Purpose of the Study:

• To apply explainable AI (XAI) for regression analysis of traffic noise levels.
• To compare the performance of ML models including KNN, XGBoost, LSTM, and RF for traffic noise prediction.
• To identify key predictors of traffic noise in urban environments.

Main Methods:

• Utilized a comprehensive traffic dataset from Dhanbad city, including vehicle speed and type.
• Implemented and compared four ML models: K-Nearest Neighbor (KNN), Extreme Gradient Boosting (XGBoost), Long-Short Term Memory (LSTM), and Random Forest (RF).
• Developed an XAI model based on the best-performing RF regressor.

Main Results:

• The Random Forest (RF) model achieved the highest performance with an RMSE of 1.27 and R-squared of 0.94.
• The XAI model, built on RF, provided interpretable insights into noise prediction.
• The number of two-wheeler vehicles was identified as a significant predictor of traffic noise levels.

Conclusions:

• Explainable AI enhances the interpretability of ML models for traffic noise prediction.
• The findings can support urban planners in developing strategies to mitigate noise pollution.
• The study highlights the effectiveness of ML and XAI in understanding and managing urban environmental factors.