Pollutant specific optimal deep learning and statistical model building for air quality forecasting.

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Poor air quality is a growing global concern with significant health impacts.
• Accurate forecasting of air pollutants is essential for environmental policy and public health interventions.
• Existing models often lack specificity for individual pollutants and long-term forecasting.

Purpose of the Study:

• To develop pollutant-specific optimal forecasting models for long-term air quality prediction.
• To investigate and compare the performance of various deep learning and statistical models.
• To provide a robust methodology for air pollutant forecasting.

Main Methods:

• Investigated eight forecasting models: stacked LSTM, LSTM auto-encoder, Bi-LSTM, convLSTM, Holt-Winters, auto-regressive (AR), SARIMA, and Prophet.
• Developed pollutant-specific models from data preprocessing through to model building.
• Utilized real-world air quality monitoring data from Kolkata over a 4-year period.

Main Results:

• Holt-Winters, Bi-LSTM, and ConvLSTM models demonstrated high forecasting accuracy for most pollutants.
• Performance was evaluated using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).
• The study identified optimal models tailored to specific air pollutants.

Conclusions:

• Pollutant-specific modeling significantly enhances long-term air quality forecasting accuracy.
• Selected models like Holt-Winters, Bi-LSTM, and ConvLSTM are effective for predicting key air pollutants.
• The findings support the development of targeted environmental policies and public health strategies.