• Version
• Download 19
• File Size 507.21 KB
• File Count 1
• Create Date 5 December 2025
• Last Updated 5 December 2025

Machine Learning for Air Quality Forecasting and Prediction- A Review of Explainable AI Methods for Enhanced Interpretability and Transparency

 

 

[1] Suraj Mohite, [2] Gauri Chavan

[1] Student, Guru Nanak Khalsa College, Mumbai, India

[2] Assistant Professor, Guru Nanak Khalsa College, Mumbai & Research Scholar, PAHER University,

Udaipur, Rajasthan, India

Corresponding Author’s Email: [1]g24.suraj.mohite@gnkhalsa.edu.in

 

 

Abstract— Air-quality forecasting is essential for public health and urban planning because pollutants such as PM2.5 and ozone have major adverse effects. Recent progress in machine learning and deep learning has substantially enhanced the precision of air-quality predictions, but their complexity often hides how predictions are formed. This review surveys peer-reviewed and preprint work published between 2019 and 2025, retrieved from open access repositories and publisher platforms, and focuses on methods that improve model interpretability. From an initial pool of records we screened, 52 studies met predefined inclusion criteria and were analysed in detail. Tree ensembles (for example Random Forest and gradient-boosted models) and deep networks (including LSTM and CNN variants) dominate forecasting experiments; alongside these, post-hoc explainability tools such as SHAP and LIME are increasingly applied to expose driver variables. We summarize the main strengths and limitations of current XAI practices in air-quality forecasting and outline priorities for method validation, reporting standards, and operational deployment.

Keywords— Air Quality Forecasting, Explainable Artificial Intelligence, Machine Learning, Model Interpretability, Pollution Prediction

 

Download