A Systematic Literature Review on the Estimation of High Air Pollution Periods Using Machine Learning Approaches

Abstract

Air pollution remains a pressing global issue, presenting serious threats to environmental sustainability and public health. This systematic review examines the application of machine learning (ML) techniques for predicting periods of elevated air pollution between 2016 and 2023, guided by the preferred reporting items for systematic reviews and meta-analyses (PRISMA) framework. The review encompasses a range of ML approaches-including classical algorithms, deep learning, ensemble methods, and hybrid models-evaluating their performance in improving prediction accuracy. Particular attention is given to the role of key pollutants, such as nitrogen oxides (NOx), and localized emission sources in influencing model outputs. While deep learning techniques effectively capture complex temporal patterns, ensemble and hybrid models demonstrate superior robustness and adaptability, especially in modeling spatial heterogeneity. Despite these advancements, notable challenges persist, including high computational requirements, limited generalizability across regions, and issues with data quality. By tracing the progression of ML applications in this field, the review synthesizes current achievements, outlines existing limitations, and offers strategic recommendations to enhance model scalability, interpretability, and practical deployment. The findings aim to guide researchers, policymakers, and environmental practitioners in advancing accurate and actionable air pollution forecasting through machine learning.