Assessment of noise pollution-prone areas using an explainable geospatial artificial intelligence approach

Area of Science:

• Environmental Science
• Geospatial Artificial Intelligence (GeoAI)
• Machine Learning

Background:

• Noise pollution is a significant environmental concern in urban areas.
• Accurate spatial assessment of noise pollution is crucial for effective urban planning and public health.
• Existing methods may lack the precision required for detailed noise pollution mapping.

Purpose of the Study:

• To spatially assess and map noise pollution prone areas in Tehran city, Iran.
• To compare the performance of different GeoAI models for noise pollution assessment.
• To identify key factors contributing to noise pollution in urban environments.

Main Methods:

• Utilized geospatial artificial intelligence (GeoAI) with the categorical boosting (CatBoost) machine learning model.
• Integrated metaheuristic algorithms: firefly algorithm (CatBoost-FA) and fruit fly optimization algorithm (CatBoost-FOA).
• Developed a spatial database including Leq, land use, traffic volume, population density, and NDVI; employed explainable AI (XAI) using SHAP for model interpretation.

Main Results:

• The CatBoost-FA model exhibited superior predictive accuracy, outperforming CatBoost-FOA and CatBoost.
• CatBoost-FA achieved the lowest RMSE (0.159 training, 0.437 test) and MAE (0.114 training, 0.371 test).
• ROC analysis showed highest accuracy for CatBoost-FA (0.897), followed by CatBoost-FOA (0.871) and CatBoost (0.846).
• SHAP analysis identified airport, commercial, and administrative zones as significant contributors to noise pollution.

Conclusions:

• GeoAI models, particularly CatBoost-FA, are effective tools for mapping urban noise pollution.
• The study highlights the importance of land use, traffic, and population density in noise pollution.
• Findings provide valuable insights for urban planners to mitigate noise pollution in Tehran.