Machine learning for environmental justice: Dissecting an algorithmic approach to predict drinking water quality in...

Area of Science:

• Environmental science
• Environmental justice
• Data science

Background:

• Machine learning (ML) offers potential for environmental monitoring and regulatory enforcement.
• Algorithmic bias can perpetuate and worsen existing societal discrimination and inequalities.

Purpose of the Study:

• To investigate the potential for embedded bias in ML algorithms used for environmental science.
• To examine how modeling decisions influence predictive outcomes and their fairness across demographic groups.

Main Methods:

• A two-part study: first, a case study on predicting drinking water quality using ML.
• Second, a dissection of algorithmic choices and their impact on model performance and fairness.

Main Results:

• ML models demonstrated varying performance in predicting drinking water quality, with some achieving over 90% accuracy.
• Algorithmic decisions significantly altered predictive outcomes and disproportionately affected the demographic characteristics of false negatives.

Conclusions:

• Vetting ML algorithms for bias is essential for environmental science and justice applications.
• Researchers and policymakers must adopt rigorous practices to ensure fairness in ML-driven environmental management.