Machine learning algorithms to forecast air quality: a survey

Area of Science:

• Environmental Science
• Computer Science
• Public Health

Background:

• Air pollution poses significant health risks, necessitating effective forecasting.
• Timely air quality predictions enable authorities to implement preventative measures.
• Machine Learning (ML), especially Deep Learning (DL), shows promise for air quality forecasting.

Purpose of the Study:

• To provide a comprehensive review of ML-based air quality forecasting.
• To analyze trends and methodologies in the field from 2011-2021.
• To classify existing research based on key parameters.

Main Methods:

• Systematic literature search across major scientific databases.
• Selection and analysis of 155 relevant research papers published between 2011 and 2021.
• Classification of papers by geographical distribution, predicted pollutants, input variables, evaluation metrics, and ML models used.

Main Results:

• The review categorizes 155 papers based on geographical scope, predicted air quality indicators, and predictor variables.
• Analysis covers diverse evaluation metrics and a wide range of Machine Learning models applied to air quality forecasting.
• Identifies key trends and common practices in the field over the decade.

Conclusions:

• Machine Learning, particularly Deep Learning, is a rapidly advancing field for air quality prediction.
• This review offers a structured overview of ML applications, aiding future research and policy development.
• Understanding current methodologies is vital for improving air quality management strategies.