Machine learning for air quality index (AQI) forecasting: shallow learning or deep learning?

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Science

Background:

• Air pollution poses significant risks to public health and the environment.
• Accurate prediction and classification of air quality are essential for effective mitigation strategies.
• Machine learning offers powerful tools for analyzing complex environmental data.

Purpose of the Study:

• To evaluate and compare various shallow learning (SL) and deep learning (DL) models for air quality index (AQI) prediction and classification.
• To identify the most effective ML models for forecasting air pollution using PM10 concentrations and meteorological data.
• To assess model performance using metrics such as accuracy, F1 score, precision, recall, and AUC.

Main Methods:

• Employed a range of ML models including Random Forest, KNN, SVM, ANN, LSTM, GRU, RNN, and CNN.
• Utilized daily PM10 concentration and nine meteorological parameters from Zabol (March 2013-February 2022).
• Applied the Information Gain (IG) method for feature selection and computed multiple performance metrics for model evaluation.

Main Results:

• Deep learning models, particularly CNN, demonstrated superior performance in AQI prediction and classification.
• CNN achieved the highest accuracy (0.60), followed closely by RF (0.58).
• DL models achieved high AUC values for classifying air quality levels (e.g., 0.95 for 'good', 0.90 for 'hazardous'), significantly outperforming SL models.

Conclusions:

• Deep learning models, especially CNN, are highly effective for complex air quality classification and prediction.
• The findings provide valuable insights for developing targeted air pollution control strategies.
• Consistent and regular air quality data collection is vital for enhancing the reliability of predictive models.