Air quality index prediction using machine learning regression models: A comparative analysis

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Not reported in abstract — specific MAE, RMSE, and R² values for each model and city are absent

• Primary Outcome: AQI prediction accuracy for Karachi and Peshawar using four ML regression models

• Nnt Or Sensitivity: Not applicable to this study type; no sensitivity, specificity, or equivalent diagnostic accuracy metrics reported; no NNT calculable

• Confidence Interval: Not reported

Clinical Application

Feasibility of deployment is not assessed. Real-time AQI prediction would require continuous sensor networks, data pipelines, and model maintenance infrastructure — none of which are discussed. The models described are static and their performance under prospective real-world conditions is unknown. Direct applicability to Australian practice is minimal. Australia operates under the National Environment Protection (Ambient Air Quality) Measure (NEPM AAQ), administered by state EPAs, with AQI monitoring conducted through established ground-station networks (e.g., NSW EPA AirWatch, EPA Victoria). Australian cities have substantially different pollutant profiles, lower baseline AQI levels, and different dominant pollution sources (bushfire smoke, traffic, industrial) compared to Karachi and Peshawar. The RACGP does not currently issue guidelines on ML-based AQI prediction tools. The TGA does not regulate environmental monitoring algorithms. PBS implications are not relevant. Clinicians advising patients with respiratory or cardiovascular conditions on air quality risk should continue to reference state EPA AQI portals rather than models derived from this study. Potentially applicable to environmental health monitoring agencies and public health authorities in high-pollution urban settings, particularly in South and Southeast Asia. Not directly applicable to individual patient clinical decision-making.

Abstract

Air plays a vital role in human life, and poor air quality can lead to respiratory infections. Given the significant impact of air quality on people's health, monitoring and assessing air quality is crucial. With advancements in machine learning (ML) and artificial intelligence (AI), we now have extensive tools to measure the Air Quality Index (AQI). Air quality is influenced by various pollutants, including carbon monoxide (CO), nitrogen dioxide (NO2), ozone (O3), and sulfur dioxide (SO2), which are prevalent in highly polluted areas and contribute to a wide range of illnesses. Particulate matter, such as PM2.5 (particles with an aerodynamic diameter of less than 2.5 µm) and PM10, poses additional health risks. To address these concerns, this study focuses on predicting AQI values for major cities in Pakistan, specifically Karachi and Peshawar, using four prominent ML algorithms: Random Forest (RF), Gradient Boosting (GB), Linear Regression (LR), and Ridge Regression (RR). The results indicate that the models effectively predicted AQI using evaluation metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the Coefficient of Determination (R2). This research's novelty lies in using the latest AQI datasets for Karachi and Peshawar and applying standard scaling for AQI normalization. Additionally, the study compares evaluation metric results across different cities, highlighting the importance of using standard scalers to achieve optimal model performance. This research underscores the value of advanced ML techniques for accurate AQI prediction and analysis.

References

1. 1.Majeed, F., Saleha, S., Abbas, L., Ghalib, S. A., Hamid, M., Saleem, M., Aman, M., & Ahmad, A. (2026). Air quality index prediction using machine learning regression models: A comparative analysis. PLoS ONE. https://doi.org/10.1371/journal.pone.0349858