Comparative Evaluation of Machine Learning Models for Short-Term Air Quality Index Prediction Using Historical Pollutant and Temporal Features in Srinagar, Jammu and Kashmir, India

This study presented a comparative machine-learning framework for Air Quality Index prediction based on historical pollutant and temporal information. An initial dataset of 7,420 observations was subjected to pollutant selection, missing-value treatment, AQI calculation, and invalid-record removal, resulting in 6,262 observations for the primary analysis. Five pollutants, PM2.5, PM10, SO2, CO, and O3, were used as the core environmental predictors. AQI was calculated using pollutant-specific sub-indices and a maximum-sub-index framework, followed by classification into six AQI categories. Two feature configurations were evaluated. The Pollution-History Model contained 86 predictors representing current pollutant conditions, pollutant history, historical AQI, pollutant changes, and rolling statistics. The Pollution + Temporal Model expanded this representation to 95 predictors by incorporating calendar and cyclical temporal variables. Six classification algorithms and five regression algorithms were evaluated using chronological time-series validation. The classification results demonstrated particularly strong performance from tree-based and ensemble methods. The Decision Tree produced a reported accuracy of 96.61% and weighted F1-score of 95.61% in Experiment B, while Bagging produced 96.55% accuracy and 95.55% weighted F1. The persistence classification baseline achieved 89.50% accuracy. For direct numerical AQI prediction, the persistence baseline achieved MAE = 14.6852, RMSE = 51.1176, and R² = 0.4581. The machine-learning regression models produced lower R² values, ranging from approximately 0.3070 to 0.3388. These results demonstrate the importance of including a simple persistence benchmark when evaluating short-term AQI prediction systems. Feature-importance analysis consistently showed that PM2.5 was the most influential predictor in the reported tree-based regression models, followed by PM10 and features representing recent pollutant changes, historical concentrations, rolling statistics, and historical AQI. The study demonstrates that historical pollutant information provides substantial predictive information for AQI-category modeling and that classical machine-learning approaches can provide strong categorical predictions. At the same time, the results show that numerical AQI prediction is more difficult and that temporal persistence represents a strong benchmark. The findings provide a foundation for subsequent research involving longer observation periods, additional environmental variables, advanced time-series models, and explainable artificial intelligence.  

 1. M. Méndez, M. G. Merayo, and M. Núñez, “Machine learning algorithms to forecast air quality: a survey,” Artificial Intelligence Review, vol. 56, pp. 10031–10066, 2023. doi: 10.1007/s10462-023-10424-4. 2. H. Liu, G. Yan, Z. Duan, and C. Chen, “Intelligent modeling strategies for forecasting air quality time series: A review,” Applied Soft Computing, vol. 102, p. 106957, 2021. doi: 10.1016/j.asoc.2020.106957. 3. D. Tang, Y. Zhan, and F. Yang, “A review of machine learning for modeling air quality: Overlooked but important issues,” Atmospheric Research, vol. 300, p. 107261, 2024. doi: 10.1016/j.atmosres.2024.107261. 4. Central Pollution Control Board, Government of India, National Air Quality Index, CPCB, New Delhi 5. N. N. Maltare and S. Vahora, “Air Quality Index prediction using machine learning for Ahmedabad city,” Digital Chemical Engineering, vol. 7, p. 100093, 2023. doi: 10.1016/j.dche.2023.100093. 6. K. Kumar and B. P. Pande, “Air pollution prediction with machine learning: a case study of Indian cities,” International Journal of Environmental Science and Technology, vol. 20, no. 5, pp. 5333–5348, 2023. doi: 10.1007/s13762-022-04241-5. 7. S. K. Natarajan, P. Shanmurthy, D. Arockiam, B. Balusamy, and S. Selvarajan, “Optimized machine learning model for air quality index prediction in major cities in India,” Scientific Reports, vol. 14, p. 6795, 2024. doi: 10.1038/s41598-024-54807-1. 8. M. T. Udristioiu, Y. El Mghouchi, and H. Yildizhan, “Prediction, modelling, and forecasting of PM and AQI using hybrid machine learning,” Journal of Cleaner Production, vol. 421, p. 138496, 2023. doi: 10.1016/j.jclepro.2023.138496. 9. Z. Wang, R. Gao, P. Wang, and H. Chen, “A new perspective on air quality index time series forecasting: A ternary interval decomposition ensemble learning paradigm,” Technological Forecasting and Social Change, vol. 191, p. 122504, 2023. doi: 10.1016/j.techfore.2023.122504. 10. Z. Gao, K. Do, Z. Li, X. Jiang, K. J. Maji, C. E. Ivey, and A. G. Russell, “Predicting PM2.5 levels and exceedance days using machine learning methods,” Atmospheric Environment, vol. 323, p. 120396, 2024. doi: 10.1016/j.atmosenv.2024.120396.