Machine Learning in Environmental Research: Common Pitfalls and Best Practices.

Area of Science:

• Environmental Science
• Data Science
• Computational Science

Background:

• Machine learning (ML) is widely adopted in environmental research for analyzing large datasets and complex variable relationships.
• Inadequate ML studies, stemming from unfamiliarity and lack of methodological rigor, can lead to erroneous conclusions in environmental science.

Purpose of the Study:

• To synthesize literature analysis and expert experience into a tutorial-like guide for environmental ML research.
• To identify and address common pitfalls in ML application within environmental studies.
• To provide best practice guidelines for rigorous data preprocessing and model development in environmental ML.

Main Methods:

• Literature analysis of 148 highly cited environmental ML research articles.
• Identification of over 30 key pitfalls and misconceptions.
• Evidence-based data analysis focusing on terminology, sample/feature size, data enrichment, feature selection, randomness assessment, data leakage, data splitting, method selection, model optimization, evaluation, explainability, and causality.

Main Results:

• Identified and categorized common misconceptions and pitfalls in environmental ML research.
• Provided evidence-based analysis and best practice recommendations for over 30 critical items.
• Illustrated good examples of supervised learning and reference modeling paradigms.

Conclusions:

• Researchers need to adopt more rigorous standards in data preprocessing and model development for environmental ML.
• Adherence to best practices ensures more accurate, robust, and practicable ML models in environmental research and applications.
• This guide aims to enhance the quality and reliability of ML studies in environmental science.