The PM 2.5 Prediction & Air Quality Classification Using Machine Learning


  • Pichitpong Soontornpipit
  • Lertsak Lekawat
  • Chatchai Tritham
  • Chattabhorn Tritham
  • Pornanong Pongpaibool
  • Narachata Prasertsuk
  • Wachirapong Jirakitpuwapat King Mongkut’s University of Technology North Bangkok (KMUTNB) Rayong Campus


air quality classification, forecast, LSTM, machine learning, PM2.5


Forecasting plays a vital role in air pollution alerts and the management of air quality. Studies and observations conducted in Thailand indicate a concerning rise in pollution levels, particularly in the concentration of PM2.5. Bangkok, in particular, has been flagged for its alarmingly high PM2.5 concentrations. By projecting the future PM2.5 concentrations in these urban areas, we can obtain valuable short-term predictive information regarding air quality. After conducting experiments using four different machine learning algorithms, it was found that the LSTM (Long Short-Term Memory) model provides the most accurate forecasts based on various statistical evaluation indicators. These indicators include a Root Mean Square Error (RMSE) of 2.74, Mean Absolute Error (MAE) of 1.97, R-squared value of 0.94, and Mean Absolute Percentage Error (MAPE) of 10.53. Then the classified air quality based on PM2.5 from the LSTM model gives the best performance indicators including accuracy = 0.9072, precision = 0.8466, negative predict value = 0.9403, sensitivity = 0.8144, specificity = 0.9381, and F1-score = 0.8169. The results show that the machine learning model can predict PM2.5 concentration, which is suitable for early warning of pollution and information provision for air quality management systems in Bangkok.




How to Cite

Soontornpipit, P., Lekawat, L., Tritham, C., Tritham, C., Pongpaibool, P., Prasertsuk, N., & Jirakitpuwapat, W. (2024). The PM 2.5 Prediction & Air Quality Classification Using Machine Learning. Thai Journal of Mathematics, 22(2), 441–452. Retrieved from