evalPM: a framework for evaluating machine learning models for particulate matter prediction.

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Particulate matter (PM) air pollution is a major global health concern.
• Accurate prediction of future PM concentrations is crucial for developing effective mitigation strategies.
• Existing machine learning (ML) models for PM prediction lack standardized evaluation, hindering comparison and selection.

Purpose of the Study:

• To introduce evalPM, a flexible framework for building, evaluating, and comparing ML models for PM prediction.
• To address the challenge of comparing diverse ML models due to varied datasets and evaluation metrics.
• To facilitate the selection of optimal ML models for specific PM immission prediction tasks.

Main Methods:

• Developed a modular framework, evalPM, offering flexibility in data sets, features, target variables, model types, hyperparameters, and evaluation metrics.
• Implemented 16 diverse ML models from existing literature within the evalPM framework.
• Conducted temporal prediction of PM concentrations using four distinct European datasets.

Main Results:

• Demonstrated the capabilities and advantages of the evalPM framework through comparative analysis of 16 ML models.
• Showcased the framework's ability to enable rapid creation and evaluation of ML-based PM prediction models.
• Highlighted the framework's utility in assessing model performance across different datasets and prediction tasks.

Conclusions:

• The evalPM framework significantly streamlines the process of developing and assessing ML models for PM prediction.
• It provides a standardized and flexible approach, overcoming limitations of previous comparative studies.
• EvalPM supports informed decision-making for selecting the most suitable ML models for environmental PM forecasting.