Fine Particulate Matter (PM2.5) Concentration Prediction in an Urban Area of Thailand Using Deep Learning Models

Fine particulate matter with an aerodynamic diameter of 2.5 micrometres or less (PM_2.5) poses major health risks. This study compares twelve deep-learning architectures — nine recurrent variants plus CNN-LSTM, PatchTST, and a Transformer — to machine-learning, statistical, and lag-1 persistence baselines for one-hour-ahead PM_2.5 forecasting at a Bangkok station (PCD-Air4Thai 35T, Lat Krabang), using 24,840 hourly observations (November 2022 – August 2025). Every deep model used one identical, leakage-free pipeline; therefore, differences reflect architecture, not preprocessing. Accuracy was assessed using RMSE, MAE, sMAPE, R², and MASE over five runs; all 171 model pairs were tested with the Diebold–Mariano test and the Benjamini–Hochberg correction ( = 0.05). The Bidirectional group (Bi-RNN, Bi-LSTM, Bi-GRU) formed the statistically best tier (hold-out RMSE = 5.712–5.743 /m³, R² = 0.652–0.656), significantly outperforming all other architectures (pBH < 0.05), with no within-group difference. The best model depended on the evaluation protocol: Bi-RNN led on the hold-out (5.712 /m³) and Bi-GRU led on the pooled RMSE (5.39 /m³). The same two-tier separation was reproduced by an expanding-window walk-forward validation (Spearman  = 0.573). Attention did not improve accuracy: all three attention variants exhibited severe cross-validation instability, and PatchTST and the Transformer performed worst. Tree-based baselines were competitive (Random Forest, 5.825 /m³), and one-step ARIMA/SARIMA were reasonable (SARIMA R² = 0.622). All models outperformed lag-1 persistence (6.258 /m³) by only ~5–9%; SHAP and LIME identified PM_2.5-lag1 as the dominant predictor. Bidirectional recurrence — not attention or transformer complexity — offers the best short-term urban PM_2.5 forecasting at this data scale.

Attention mechanism; Deep learning; PM2.5 forecasting; Recurrent neural network; Time series analysis

Ahmad, F. A., Liu, J., Hashim, F., & Samsudin, K. (2024). Short-term load forecasting utilizing a combination model: A brief review. International Journal of Technology, 15(1), 121–129. https://doi.org/10.14716/ijtech.v15i1.5543

Al-Selwi, H. F., Abd. Aziz, A., Bin Abas, F., Kayani, A., & Noor, N. M. (2023). Attention based spatial-temporal GCN with Kalman filter for traffic flow prediction. International Journal of Technology, 14(6), 1299–1308. https://doi.org/10.14716/ijtech.v14i6.6646

Bahdanau, D., Cho, K., & Bengio, Y. (2015). Neural machine translation by jointly learning to align and translate. Proceedings of the 3rd International Conference on Learning Representations (ICLR 2015). https://arxiv.org/abs/1409.0473

Bergmeir, C., & Ben´?tez, J. M. (2012). On the use of cross-validation for time series predictor evaluation. Information Sciences, 191, 192–213. https://doi.org/10.1016/j.ins.2011.12.028

Bhatta, J., Acharya, S. R., & Yang, K. M. (2026). Machine learning-enhanced air quality forecasting and trend analysis: A five-year comprehensive assessment of PM2.5 concentrations in Bangkok, Thailand. Environmental Challenges, 22, 101442. https://doi.org/10.1016/j.envc.2026.101442

 Box, G. E. P., Jenkins, G. M., Reinsel, G. C., & Ljung, G. M. (2015). Time series analysis: Forecasting and control (5th). John Wiley & Sons.

Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324

Cho, K., van Merri¨enboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014). Learning phrase representations using RNN encoder–decoder for statistical machine translation. Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), 1724–1734. https://doi.org/10.3115/v1/D14-1179

Cohen, A. J., Brauer, M., Burnett, R., Anderson, H. R., Frostad, J., Estep, K., & Forouzanfar, M. H. (2017). Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: An analysis of data from the Global Burden of Diseases Study 2015. The Lancet, 389(10082), 1907–1918. https://doi.org/10.1016/S0140-6736(17)30505-6

Diebold, F. X., & Mariano, R. S. (1995). Comparing predictive accuracy. Journal of Business & Economic Statistics, 13(3), 253–263. https://doi.org/10.1080/07350015.1995.10524599

Drewil, G. I., & Al-Bahadili, R. J. (2022). Air pollution prediction using LSTM deep learning and metaheuristics algorithms. Measurement: Sensors, 24, 100546. https://doi.org/10.1016/j.measen.2022.100546

Duan, L.-C., Minh, N.-N., Dung, T.-C., & Linh, T.-T.-T. (2026). Energy-efficient TinyML approach for wearable fall detection on edge devices using spatial-temporal deep learning. International Journal of Technology, 17(3), 919–935. https://doi.org/10.14716/ijtech.v17i3.8445

Elman, J. L. (1990). Finding structure in time. Cognitive Science, 14(2), 179–211. https://doi.org/10.1207/s15516709cog1402_1

Eren, B., Erden, C., Atal?, A., & Ozdemir, S. (2025). A comparative analysis of hyperparameter optimization using LSTM-based deep learning models for urban air quality predictions. Ain Shams Engineering Journal, 16, 103786. https://doi.org/10.1016/j.asej.2025.103786

Escanciano, J. C., & Parra, R. (2026). Extending the scope of inference about predictive ability to machine learning methods [Published online: 14 January 2026]. Journal of Business & Economic Statistics, 1–12. https://doi.org/10.1080/07350015.2025.2562964

Friedman, J. H. (2001). Greedy function approximation: A gradient boosting machine. Annals of Statistics, 29(5), 1189–1232. https://doi.org/10.1214/aos/1013203451

Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press. https://www.deeplearningbook.org

Gupta, V., Dash, Y., Goyal, A., Lodh, A., Singh, V., Routray, A., & Kumar, P. (2026). A BiLSTM-driven framework for operational PM2.5 forecasting: Integrating meteorological kinematics for urban air quality management. Frontiers in Climate, 8, 1855755. https://doi.org/10.3389/fclim.2026.1855755

Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735

Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd). OTexts. https://otexts.com/fpp3

IQAir. (2024). World air quality report 2023 (tech. rep.). IQAir. https://www.iqair.com/world-air-quality-report

Keerthana, G., & Ushasree, R. (2024). Air quality prediction using RNN and LSTM. Journal of Innovation and Technology, 2024(48).

LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. https://doi.org/10.1038/nature14539

Lundberg, S. M., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems (NeurIPS), 30, 4765–4774.

Makridakis, S., Spiliotis, E., & Assimakopoulos, V. (2018). Statistical and machine learning forecasting methods: Concerns and ways forward. PLOS ONE, 13(3), e0194889. https://doi.org/10.1371/journal.pone.0194889

Park, S.-Y., Woo, S.-H., & Lim, C. (2023). Predicting PM10 and PM2.5 concentration in container ports: A deep learning approach. Transportation Research Part D, 115, 103601. https://doi.org/10.1016/j.trd.2022.103601

Parntrasri, W., & Puangthongthub, S. (2026). Weekly extremes of PM2.5-attributable cardiovascular mortality in Northeastern Thailand: Extreme-value return levels and 26-week forecasts. City and Environment Interactions, 29, 100313. https://doi.org/10.1016/j.cacint.2026.100313

Patel, P., Patel, S., Shah, K., Desai, K., Patel, S., Shah, M., & Patel, S. (2025). A systematic study on PM2.5 and PM10 concentration prediction in air pollution using machine learning and deep learning model. Environmental Chemistry and Ecotoxicology, 7, 1401–1415. https://doi.org/10.1016/j.enceco.2025.07.001

Pope, C. A., & Dockery, D. W. (2006). Health effects of fine particulate air pollution: Lines that connect. Journal of the Air & Waste Management Association, 56(6), 709–742. https://doi.org/10.1080/10473289.2006.10464485

Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). “why should I trust you?”: Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1135–1144. https://doi.org/10.1145/2939672.2939778

Schmidhuber, J. (2015). Deep learning in neural networks: An overview. Neural Networks, 61, 85–117. https://doi.org/10.1016/j.neunet.2014.09.003

Schuster, M., & Paliwal, K. K. (1997). Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11), 2673–2681. https://doi.org/10.1109/78.650093

Sharma, E., Deo, R. C., Prasad, R., Parisi, A. V., & Raj, N. (2020). Deep air quality forecasts: Suspended particulate matter modeling with convolutional neural and long short-term memory networks. IEEE Access, 8, 244965. https://doi.org/10.1109/ACCESS.2020.3039002

Son, R., Stratoulias, D., Kim, H. C., & Yoon, J.-H. (2023). Estimation of surface PM2.5 concentrations from atmospheric gas species retrieved from TROPOMI using deep learning: Impacts of fire on air pollution over Thailand. Atmospheric Pollution Research, 14, 101875. https://doi.org/10.1016/j.atmosres.2023.101875

Sreenivasulu, T., & Mokesh Rayalu, G. (2025). Accurate hourly AQI prediction using temporal CNN-LSTM-MHA+GRU: A case study of seasonal variations and pollution extremes in Visakhapatnam, India. Results in Engineering, 27, 106303. https://doi.org/10.1016/j.rineng.2025.106303

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems (NeurIPS), 30.

Wang, X., Liu, X., & Mao, W. (2026). Air quality estimation from sequential surveillance images using a unified CNN-RNN framework. iScience, 29, 114919. https://doi.org/10.1016/j.isci.2026.114919

WHO. (2021). WHO global air quality guidelines: Particulate matter (PM2.5 and PM10), ozone, nitrogen dioxide, sulfur dioxide and carbon monoxide (tech. rep.). World Health Organization. https://www.who.int/publications/i/item/9789240034228

Zhang, X., & Zhou, P. (2024). A transferred spatio-temporal deep model based on multi-LSTM auto-encoder for air pollution time series missing value imputation. Future Generation Computer Systems, 156, 325–338. https://doi.org/10.1016/j.future.2024.03.015

Zhao, Z., Qin, J., Hu, A., Zhang, N., Xie, J., & Sun, Y. (2026). WaveNet-LSTM: A deep learning air quality prediction model based on both air pollutants and meteorological factors. Information Sciences, 730, 122894. https://doi.org/10.1016/j.ins.2025.122894