Comparative Analysis Study of Air Quality Index Prediction Using Regression and Deep Learning Techniques

We set out to benchmark nine models — spanning classical, ensemble, and deep learning families — for AQI point estimation and 24-hour forecasting, using 74,682 hourly records across six Indian cities under realistic sensor-fault conditions. Random Forest came out on top for point estimation, though as discussed above, this largely reflects its ability to learn the deterministic CPCB formula rather than a harder predictive task; LSTM remains the better choice for genuine multi-step forecasting. Non-linear models clearly outperformed the linear baselines, confirming that cross-pollutant interactions matter, and SVR's poor showing with default hyperparameters is a reminder that it needs careful tuning before deployment. Overall, this work offers a reproducible benchmark and a starting framework for smart-city AQI monitoring — one that still needs validation on real station data before its conclusions can be generalised.

1.      WHO, 'Ambient Air Quality Database,' World Health Organization, Geneva, 2021. 2.      CPCB, 'National Air Quality Index: Technical Manual,' Ministry of Environment, Forest and Climate Change, Govt. of India, New Delhi, 2014. 3.      IQAir, 'World Air Quality Report 2023: Region & City PM2.5 Ranking,' IQAir Foundation, Goldach, Switzerland, 2024. 4.      UNEP, 'Air Pollution in Asia and the Pacific: Science-Based Solutions,' UNEP Regional Office for Asia and the Pacific, Bangkok, 2023. 5.      Z. Zhang, C. Wu, J. Luo, and Y. Xie: 'Hybrid ARIMA-LSTM framework for nationwide PM2.5 forecasting in China,' Atmospheric Environment, 2022, 278, p. 119085. 6.      A. Ali, M. Zafar, and I. Shah: 'Regression-based air quality prediction using PEPA monitoring network data in Pakistan,' Environmental Monitoring and Assessment, 2021, 193, (6), pp. 1–18. 7.      D. Pandey, N. Kumar, and R. Singh: 'Random Forest-based AQI prediction for Indian metropolitan cities using CPCB data,' Journal of Environmental Informatics Letters, 2022, 8, (2), pp. 112–124. 8.      Y. Huang, T. Lin, J. Zheng, and H. Li: 'XGBoost-GRU hybrid model for PM2.5 hourly forecasting in Shanghai,' Science of the Total Environment, 2022, 844, p. 157008. 9.      S. Mohan, R. Rao, and K. Bhatt: 'Gradient boosting ensemble for urban AQI prediction in Mumbai metropolitan region,' Sustainable Computing: Informatics and Systems, 2023, 38, p. 100847. 10.   K. Gu, X. Zhao, B. Qiao, and J. Qin: 'LSTM with attention mechanism for PM2.5 concentration forecasting in Beijing,' IEEE Transactions on Neural Networks and Learning Systems, 2023, 34, (8), pp. 4821–4833.