Air Quality Index Prediction with Long Short‐Term Memory—Moving Average (LSTM‐MA)—Machine Learning Model

Air pollution is a significant factor in public health challenges, increasing the rates of respiratory and heart-related illnesses. Accurate forecasting of changes in air quality is essential for raising community awareness and managing pollution effectively in urban areas. This research applies artificial intelligence methodologies for air quality forecasting in Ba Ria - Vung Tau Province, Vietnam. A long short-term memory (LSTM) model was employed, enhanced by a moving average method to mitigate dataset noise. An evaluation of the statistics shows that the air quality in the assessed region is predominantly satisfactory, with concentrations of some primary pollutants, such as NO2, SO2, and suspended particulate matter, remaining below the established regulatory limits. O3 centent frequently exceeded permissible thresholds, exerting a notable influence on VN_AQI. The developed LSTM-MA model demonstrated good predictive accuracy over a 7-day horizon, achieving a value of RMSE within the range of 0.157 - 0.271 and MAPE of 30 – 31%. For extended forecasts spanning 2 weeks to 1 month, the model maintained favorable performance, with RMSE and MAPE values ranging from 0.186 to 0.497 and 46.3 to 84.4, respectively. Additionally, comparative evaluations were conducted against conventional forecasting models, including LSTM and ARIMA. The findings showed that the LSTM-MA model exhibited exemplary performance and holds significant promise for predicting fluctuations in air quality over time.