Abstract

The prediction of fine particulate matter, or fine particulate (PM₂.₅), is very important for air quality management and protecting human health. Current deep-learning approaches, however, cannot capture the local time-specific interactions and spatiotemporal correlations across long time periods, or are computationally too expensive to effectively model the seasonal periodic variations. This study builds on these insights to suggest the Convolutional Attention Seasonal Aggregation Network (CASA-Net) combining Conv1D feature extraction, Bidirectional Long Short-Term Memory (BiLSTM), Multi-Head Self-Attention (MHSA), residual learning, and sinusoidal Day-of-Year (DOY) seasonal embeddings. The proposed model was tested against the day-by-day PM₂.₅ data for the Yorkshires provided by the United States Agency for Protection of the Environment (EPA) in the Air Quality System (AQS). The proposed model was tested using the Daily PM₂.₅ data for Yorkshires from the United States Agency for Protection of the Environment (EPA) Air Quality System (AQS) with 133,625 sliding window observations from 543 quality-controlled stations. To prevent temporal leakage and provide reliable model assessment, the block cross-validation approach with a train--validation--test split of 70:15:15, which is geographically diverse, was used. The experimental results show that CASA-Net can significantly reduce RMSE, MAE, and MAPE values from machine learning-based models and the latest deep learning baseline by getting RMSE of 3.1847 µg/m³, MAE of 1.7214 µg/m³, R² of 0.8134, MAPE of 10.84%, and a Pearson correlation coefficient (r) of 0.9041. CASA-Net outperforms Informer in terms of the RMSE reduction (18.22%) and the increase in R² value (0.1503), as well as fewer required parameters (4.9× fewer) and faster training and inference time, and estimated fewer required GFLOPs per forward pass (8.8× fewer). Moreover, the robustness, computational efficiency, and application for next-day PM (2.5) forecasting and air quality decision-making are also demonstrated by applying statistical significance testing, uncertainty quantification, ablation studies, feature attribution, and attention analysis to CASA-Net.

Keywords

Funding

This research received no external funding.

Acknowledgment

The Grammarly AI Tool is used to make grammar corrections in the Manuscript.

References

1. Bhatt, D., & Goswami, S. (2026). Air Quality Prediction and Forecasting Using Machine Learning Algorithms: A Review. DMPedia Lecture Notes in Multidisciplinary Research, 1095-1108. https://doi.org/10.65890/dmp.lnmr.IMPACT26.59
2. Nath, B., Mohamed, F. A., & Hassan, F. (2026). Towards Green AI: A Lightweight Experimental Study on Carbon-Aware Neural Network Training. Impact in Computics, 2, 1-13. https://doi.org/10.65500/computics-2026-001
3. Agbehadji, I. E., & Obagbuwa, I. C. (2024). Systematic review of machine learning and deep learning techniques for spatiotemporal air quality prediction. Atmosphere, 15(11), 1352. https://doi.org/10.3390/atmos15111352
4. Ata, O., & Onay, Ö. (2026). Frost Protection Strategies in Agriculture: Insights from the 2025 Niğde Frost Event in Türkiye. Impact in Agriculture, 2, 1-13. https://doi.org/10.65500/agriculture-2026-001
5. Zhang, Z., Zeng, Y., & Yan, K. (2021). A hybrid deep learning technology for PM2. 5 air quality forecasting. Environmental Science and Pollution Research, 28(29), 39409-39422. https://doi.org/10.1007/s11356-021-12657-8
6. Zhou, S., Wang, W., Zhu, L., Qiao, Q., & Kang, Y. (2024). Deep-learning architecture for PM2. 5 concentration prediction: A review. Environmental Science and Ecotechnology, 21, 100400. https://doi.org/10.1016/j.ese.2024.100400
7. Kim, H. S., Han, K. M., Yu, J., Youn, N., & Choi, T. (2025). Development of a Hybrid Attention Transformer for Daily PM2. 5 Predictions in Seoul. Atmosphere, 16(1), 37. https://doi.org/10.3390/atmos16010037
8. Zhang, R., & Awang, N. (2025). A novel ST-iTransformer model for spatio-temporal ambient air pollution forecasting. Journal of Big Data, 12(1), 101. https://doi.org/10.1186/s40537-025-01150-5
9. Wu, C., Wang, R., Lu, S., Tian, J., Yin, L., Wang, L., & Zheng, W. (2025). Time-series data-driven PM2. 5 forecasting: from theoretical framework to empirical analysis. Atmosphere, 16(3), 292. https://doi.org/10.3390/atmos16030292
10. Pan, P., Malarvizhi, A. S., & Yang, C. (2025). Data Augmentation Strategies for Improved PM2. 5 Forecasting Using Transformer Architectures. Atmosphere, 16(2), 127. https://doi.org/10.3390/atmos16020127
11. Cáceres-Tello, J., & Galán-Hernández, J. J. (2024). Analysis and prediction of PM2. 5 pollution in Madrid: The use of Prophet–Long Short-Term Memory hybrid models. AppliedMath, 4(4), 1428-1452. https://doi.org/10.3390/appliedmath4040076
12. Nguyen, T. N. T., & Trinh, T. D. (2026). Satellite-Driven Prediction of Fine Particulate Matter (PM2. 5) Concentrations: Machine Learning and Explainable Artificial Intelligence. Engineering Research Express. https://doi.org/10.1088/2631-8695/ae7028
13. Teeravech, K., Samphutthanont, R., & Limrueangrong, P. (2026, January). Machine Learning Approaches for 2025 PM2. 5 Forecasting in Thailand. In 2026 18th International Conference on Knowledge and Smart Technology (KST) (pp. 422-427). IEEE. https://doi.org/10.1109/KST67832.2026.11432409
14. Wang, Z., Jia, K., Zhang, W., & Zhang, C. (2025). PM2. 5 Concentration Prediction in the Cities of China Using Multi-Scale Feature Learning Networks and Transformer Framework. Sustainability, 17(19), 8891. https://doi.org/10.3390/su17198891
15. Muruganandam, N. S., & Arumugam, U. (2022). Seminal stacked long short-term memory (SS-LSTM) model for forecasting particulate matter (PM2. 5 and PM10). Atmosphere, 13(10), 1726. https://doi.org/10.3390/atmos13101726
16. Liu, Z., Fang, Z., & Hu, Y. (2025). A deep learning-based hybrid method for PM2. 5 prediction in central and western China. Scientific Reports, 15(1), 10080. https://doi.org/10.1038/s41598-025-95460-6
17. Khadom, A. A., Albawi, S., Abboud, A. J., Mahood, H. B., & Hassan, Q. (2024). Predicting air quality index and fine particulate matter levels in Bagdad city using advanced machine learning and deep learning techniques. Journal of Atmospheric and Solar-Terrestrial Physics, 262, 106312. https://doi.org/10.1016/j.jastp.2024.106312
18. Hu, Y. Y., Liao, H. B., Yuan, L., & Deng, Y. Z. (2024). MTLPM: a long-term fine-grained PM2. 5 prediction method based on spatio-temporal graph neural network. Environmental Monitoring and Assessment, 196(12), 1240. https://doi.org/10.1007/s10661-024-13407-2
19. Srimanee, T., Kiatphibun, B., Thanomngern, C., Mhuaksa, N., Banju-ngam, C., & Siriborvornratanakul, T. (2026). Multimodal PM2. 5 Forecasting Using Satellite Imagery and Sensor Data with Semi-supervised Deep Learning. Earth Systems and Environment, 1-18. https://doi.org/10.1007/s41748-026-01109-3
20. Latoń, D., Grela, J., Ożadowicz, A., & Wisniewski, L. (2025). Artificial intelligence and machine learning approaches for indoor air quality prediction: a comprehensive review of methods and applications. Energies, 18(19), 5194. https://doi.org/10.3390/en18195194
21. Wu, J., Xu, L., Fan, S., Xia, K., & Wang, L. (2025). A Spatiotemporal graph attention network for PM2. 5 forecasting using multi-source data. Air Quality, Atmosphere & Health, 18(10), 3037-3052. https://doi.org/10.1007/s11869-025-01818-0
22. Verma, A., Ranga, V., & Vishwakarma, D. K. (2024). BREATH-Net: a novel deep learning framework for NO2 prediction using bi-directional encoder with transformer. Environmental Monitoring and Assessment, 196(4), 340. https://doi.org/10.1007/s10661-024-12455-y
23. Hossen, M. K., Peng, Y. T., Shao, A., & Chen, M. C. (2025). An ODE based neural network approach for PM2. 5 forecasting: MK Hossen et al. Scientific Reports, 15(1), 24830. https://doi.org/10.1038/s41598-025-05958-2
24. Rosca, C. M., Carbureanu, M., & Stancu, A. (2025). Data-driven approaches for predicting and forecasting air quality in urban areas. Applied Sciences, 15(8), 4390. https://doi.org/10.3390/app15084390
25. Bai, J., Nie, W., Zhang, Z., Mehmood, M., & Dong, L. (2026). An optimized deep learning model with error correction for forecasting particulate matteR².5 concentrations near tailings ponds. International Journal of Environmental Science and Technology, 23(4), 315. https://doi.org/10.1007/s13762-026-07067-7
26. Balasubramanian, C., Kareemullah, H., Arunkumar, R. S., Kumar, P. S., & Revathi, S. M. (2026). Multi-objective deep Q-self-attention transformer for PM2. 5 prediction on edge devices. GeoInformatica, 30(1), 11. https://doi.org/10.1007/s10707-026-00565-3
27. Adduri, S. D., & Ali, T. (2025). BREATHE: A GeoAI-Powered Air Quality Monitoring and Forecasting System for Urban Sustainability. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 48, 39-46. https://doi.org/10.5194/isprs-archives-XLVIII-G-2025-39-2025
28. Wang, J., Wu, T., Mao, J., & Chen, H. (2024). A forecasting framework on fusion of spatiotemporal features for multi-station PM2.5 concentration prediction. Expert Systems with Applications, 238, 121951. https://doi.org/10.1016/j.eswa.2023.121951
29. Zhou, S., Zhang, X., Liu, J., Zhang, Y., Wei, P., Wang, Y., & Zhang, J. (2024). Long-term prediction of particulate matter (PM2.5) concentration with modal Autoformer based on fusion modal decomposition algorithm. Atmosphere, 15(1), 4. https://doi.org/10.3390/atmos15010004
30. He, Z., & Guo, Q. (2024). Comparative analysis of multiple deep learning models for forecasting monthly ambient PM2.5 concentrations: A case study in Dezhou City, China. Atmosphere, 15(12), 1432. https://doi.org/10.3390/atmos15121432
31. Xue, Y., Guan, S., & Jia, W. (2025). PMformer: A novel Informer-based model for accurate long-term time series prediction. Information Sciences, 690, 121586. https://doi.org/10.1016/j.ins.2024.121586
32. Chen, Y., Wang, J., Li, R., & Gao, J. (2025). PM2.5 concentration prediction system combining fuzzy information granulation and multi-model ensemble learning. Journal of Environmental Sciences, 156, 332–345. https://doi.org/10.1016/j.jes.2024.07.010
33. Jia, W., Guan, S., & Liu, W. (2026). Environmental forecasting of particulate matter (PM2.5) using seasonal-trend decomposition and iTransformer: A multivariate learning framework. Environmental Engineering Science. https://doi.org/10.1177/15579018251397145
34. Al-qaness, M. A. A., Dahou, A., Ewees, A. A., Abualigah, L., Huai, J., Abd Elaziz, M., & Helmi, A. M. (2023). ResInformer: Residual transformer-based artificial time-series forecasting model for PM2.5 concentration in three major Chinese cities. Mathematics, 11(2), 476. https://doi.org/10.3390/math11020476
35. Liang, Y., Xia, Y., Ke, S., Wang, Y., Wen, Q., Zhang, J., Zheng, Y., & Zimmermann, R. (2022). AirFormer: Predicting nationwide air quality in China with Transformers. https://doi.org/10.1609/aaai.v37i12.26676
36. Wu, H., Xu, J., Wang, J., & Long, M. (2021). Autoformer: Decomposition Transformers with Auto-Correlation for Long-Term Series Forecasting. Advances in Neural Information Processing Systems, 34, 22419–22430.
37. Zeng, A., Chen, M., Zhang, L., & Xu, Q. (2023). Are Transformers Effective for Time Series Forecasting? Proceedings of the AAAI Conference on Artificial Intelligence, 37(9), 11121–11128. https://doi.org/10.1609/aaai.v37i9.26317
38. Tyagi, L., Singh, D., & Goyal, N. (2026). Enhancing oil spill classification for sustainable environment: Leveraging CNNs and self-attention mechanisms. International Journal of Sustainable Building Technology and Urban Development, 17(1), 24-40. https://doi.org/10.22712/susb.20260003
39. Verma, V., Gupta, D., Gupta, S., Uppal, M., Anand, D., Ortega-Mansilla, A., ... & Goyal, N. (2022). A deep learning-based intelligent garbage detection system using an unmanned aerial vehicle. Symmetry, 14(5), 960. https://doi.org/10.3390/sym14051045