A Comparative Analysis of Machine Learning Models for Real-Time PM2.5 Prediction and Health Planning

Abstract

This paper proposes an intelligent air quality prediction system that combines real-time PM2.5 forecasting, automated geolocation, and a 7-day AI health planner within a single, integrated architecture. Unlike traditional environmental monitoring networks that rely heavily on sparse hardware sensors and delayed data updates, the proposed approach leverages machine learning methods. By taking into account spatial, temporal, chemical, and meteorological factors, the system significantly improves predictive accuracy and delivers actionable, hyper-local insights. The architecture is built on a scalable web framework that facilitates seamless communication between its modules. To ensure real-time optimization, the system integrates dynamic geocoding and live data ingestion via Open-Meteo APIs, bypassing the need for manual data entry. Through rigorous comparative analysis, the Random Forest Regressor was deployed as the core predictive engine, demonstrating superior capability in handling the highly non-linear and volatile nature of atmospheric data compared to conventional linear models. Early experimental analysis shows the proposed system optimally predicts PM2.5 levels while successfully accounting for concept drift between historical data and real-time sensor readings. The suggested framework resolves the main weaknesses of current air quality platforms, such as faulty localized forecasting, the absence of system integration, and a lack of personalized health guidance, making it highly applicable for modern public health and environmental management.

Introduction

The text discusses the growing problem of air pollution in India, especially fine particulate matter (PM2.5), and the limitations of traditional air quality monitoring systems. Existing government sensor networks are sparse, slow, and unable to provide accurate, real-time, localized forecasts. They also rely on static statistical methods that fail to capture dynamic factors like traffic, weather changes, and industrial emissions.

To address these issues, the study highlights the use of machine learning (ML) and artificial intelligence (AI) for more accurate and real-time air quality prediction. Among various models, ensemble methods like Random Forest perform best because they can handle complex, non-linear environmental data more effectively than traditional methods or simpler ML models.

The paper proposes an integrated AI-based web system that combines:

• Real-time PM2.5 prediction
• Automated geolocation and weather data fetching
• Machine learning-based forecasting (using Random Forest)
• A 7-day AI health planner that provides personalized health advice
• A user-friendly dashboard with live environmental insights

The literature review shows a shift from traditional sensor-based and statistical models to data-driven ML approaches, but also highlights a major gap: most existing models are not integrated into real-world, user-facing systems and lack actionable health guidance.

A comparative analysis of models (KNN, Decision Tree, Random Forest) shows that Random Forest performs best, with the lowest prediction error and highest accuracy, making it suitable for real-time deployment.

The proposed system architecture includes:

• Automatic location detection using browser APIs
• Real-time environmental data fetching via external APIs
• A Flask-based ML inference engine
• Data preprocessing and feature scaling
• AQI classification and health advisory generation
• A transparent UI showing both predicted values and real sensor readings

Conclusion

The rapid deterioration of urban air quality necessitates a transition from reactive environmental monitoring to proactive, predictive intelligence. This paper introduced an end-to-end, AI-powered air quality prediction system designed to provide hyper-local PM2.5 forecasting, real-time variance education, and multi-day health planning. A critical contribution of this research was the rigorous comparative analysis of machine learning algorithms for atmospheric forecasting. While traditional models like Linear Regression and standalone Decision Trees failed to manage the complex, non-linear, and highly volatile nature of environmental data, ensemble learning proved highly effective. The Random Forest Regressor emerged as the superior intelligence engine, successfully mitigating overfitting and achieving the highest predictive accuracy ($R^2$ = 0.89) by naturally capturing the intricate chemical synergies between precursor pollutants. Furthermore, this research successfully bridged the gap between theoretical machine learning and practical public health application. By embedding the Random Forest model within a scalable, microservices-based web architecture and integrating live geolocation with Open-Meteo APIs, the system eliminates manual data entry and dynamically adapts to the user\'s immediate environment. The introduction of the 7-Day AI Health Planner elevates the system from a simple data dashboard to an actionable, life-saving advisory tool.

References

[1] Breiman, L. (2001). \"Random Forests.\" Machine Learning, 45(1), 5-32. [2] Sharma, S., & Kumar, A. (2022). \"Machine Learning Techniques for Air Quality Forecasting in Urban India: A Comparative Study.\" Environmental Science and Pollution Research, 29(14), 21345-21358. [3] Rybarczyk, Y., & Zalakeviciute, R. (2018). \"Machine Learning Approaches for Outdoor Air Quality Modelling: A Systematic Review.\" Applied Sciences, 8(12), 2570. [4] Pedregosa, F., et al. (2011). \"Scikit-learn: Machine Learning in Python.\" Journal of Machine Learning Research, 12, 2825-2830. [5] Gocheva-Ilieva, S. G., Ivanov, A. V., & Voynikova, D. S. (2019). \"Predicting Daily Fine Particulate Matter (PM2.5) Concentrations Using Ensemble Learning and K-Nearest Neighbors.\" Atmospheric Pollution Research, 10(3), 856-865. [6] Gama, J., Zliobaite, I., Bifet, A., Pechenizkiy, M., & Bouchachia, A. (2014). \"A Survey on Concept Drift Adaptation.\" ACM Computing Surveys (CSUR), 46(4), 1-37. [7] Balakrishnan, K., et al. (2019). \"The Impact of Air Pollution on Deaths, Disease Burden, and Life Expectancy across the States of India: The Global Burden of Disease Study 2017.\" The Lancet Planetary Health, 3(1), e26-e39. [8] Hastie, T., Tibshirani, R., & Friedman, J. (2009). The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer Science & Business Media. [9] Xia, Y., et al. (2020). \"A Web-Based System for Air Quality Forecasting and Visualization Using Machine Learning.\" IEEE Access, 8, 123456-123465. [10] Zafra-Cabeza, A., et al. (2020). \"Integration of Real-Time API Meteorological Data for Dynamic Air Quality Index Prediction.\" Sensors, 20(18), 5123.

Copyright

Copyright © 2026 Vasu Agarwal, Utkarsh Awasthi, Suryansh Kushwaha, Vibhor Sachan, Namrata Ambekar . This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.