AI-Driven Air Quality Index Prediction and Real-Time Health Advisory System Using Machine Learning

This paper presented an end-to-end AI-driven system for Air Quality Index prediction and real-time health advisory generation, evaluated through a comparative analysis of five machine learning models on a CPCB-formula-validated dataset of 10,000 samples across eight major Indian cities. The experimental results establish a clear performance hierarchy among the evaluated models. Linear Regression, serving as the baseline, achieved R2 = 0.8347, confirming that the CPCB breakpoint formula introduces substantial nonlinearity requiring more sophisticated modeling approaches. The Decision Tree (R2 = 0.9814) and Random Forest (R2 = 0.9878) demonstrated the natural affinity of tree-based architectures for the piecewise linear structure of the AQI formula. The MLP neural network (R2 = 0.9764), while effective, underperformed relative to tree-based methods, consistent with the literature on neural networks for tabular prediction tasks [9, 10]. XGBoost achieved the highest overall performance with R2 = 0.9938, MAE = 2.40, and MAPE = 1.54%, demonstrating that regularized gradient boosting effectively captures the complex, threshold-based relationships in AQI computation. Per-category analysis revealed that model accuracy improves with pollution severity: the Hazardous category was predicted with exceptional precision (MAE = 0.56, R2 = 0.9980), while the Moderate category showed weaker performance due to limited training representation. This finding has important practical implications, the model is most accurate precisely in the conditions where accurate prediction is most critical for public health. The system's multi-tier real-time data acquisition architecture, with automatic fallback between WAQI ground-station data and OpenWeatherMap satellite estimates, addresses the practical challenge of data source reliability in operational deployment. The health advisory engine translates predicted AQI values into specific, actionable recommendations across five severity categories, bridging the gap between raw environmental measurements and public health guidance. The parallels between the ML pipeline presented in this paper and the constructive neural network frameworks of Kaur and Gupta [10, 11, 12] suggest a unifying principle: iterative, data-driven optimization from simple components to complex predictive systems is effective across different levels of computational intelligence. Combined with the secure data management frameworks proposed through blockchain integration [13, 14, 15], these approaches point toward a future of more reliable, transparent, and actionable intelligent environmental monitoring systems. Future work should investigate: (a) incorporating temporal features and LSTM/Transformer architectures for AQI time-series forecasting; (b) integrating IoT sensor networks for hyperlocal pollution monitoring; (c) developing mobile applications with push notification alerts for real-time health advisories; (d) applying model interpretability techniques (SHAP, LIME) for pollution source attribution; (e) extending the system to international air quality standards (US EPA, European EEA); and (f) conducting clinical validation of the health advisory framework through epidemiological studies.

[1] World Health Organization. "Ambient (Outdoor) Air Pollution." WHO Fact Sheet, 2022. https://www.who.int/news-room/fact-sheets/detail/ambient-(outdoor)-air-quality-and-health [2] Balakrishnan, K., Dey, S., Gupta, T., et al. "The Impact of Air Pollution on Deaths, Disease Burden, and Life Expectancy Across the States of India." The Lancet Planetary Health 3(1) (2019): e26-e39. [3] Central Pollution Control Board. "National Air Quality Index." CPCB, Ministry of Environment, Forest and Climate Change, Government of India (2014). [4] Kumar, A., and Goyal, P. "Forecasting of Air Quality in Delhi Using Principal Component Regression Technique." Atmospheric Pollution Research 2(4) (2011): 436-444. [5] Zheng, Y., Liu, F., and Hsieh, H. P. "U-Air: When Urban Air Quality Inference Meets Big Data." Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2013): 1436-1444. [6] Sharma, E., Deo, R. C., Prasad, R., and Parisi, A. V. "A Hybrid Air Quality Early-Warning Framework: An Hourly Forecasting Model with Online Sequential Extreme Learning Machine and Empirical Mode Decomposition Algorithm." Science of The Total Environment 709 (2020): 135934. [7] Breiman, L. "Random Forests." Machine Learning 45(1) (2001): 5-32. [8] Chen, T., and Guestrin, C. "XGBoost: A Scalable Tree Boosting System." Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (2016): 785-794. [9] Kaur, J., and Gupta, N. "Artificial Neural Network: A Review." International Journal of Technical Research & Science (n.d.): 1-4. [10] Kaur, J., and Gupta, N. "Constructive Neural Network: A Framework." International Journal of Engineering and Advanced Technology 9(2) (2019). https://doi.org/10.35940/ijeat.b3304.129219