Identifying schools at high-risk for elevated lead in drinking water using only publicly available data.

Area of Science:

• Environmental Science
• Public Health
• Data Science

Background:

• Lead contamination in school drinking water poses a significant health risk to children.
• Estimating this risk at a state level is challenging due to variable water quality and unknown plumbing material locations.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting the likelihood of lead contamination in school drinking water using publicly available data.
• To create a tool for state-level risk assessment to guide resource allocation for lead exposure prevention.

Main Methods:

• Six machine learning models were trained and tested using data from over 8,000 schools in California and Massachusetts.
• Predictive features included water chemistry, US census socioeconomic data, and Geographic Information System spatial data.
• A Random Forest model demonstrated the best performance, achieving high ROC AUC scores.

Main Results:

• The Random Forest model achieved 10-fold cross-validation ROC AUC scores of 0.88 for Massachusetts and 0.78 for California.
• The model successfully categorized lead leaching risk for schools, showing good agreement with actual outcomes.
• A slight overestimation of lead leaching risk was observed in up to 17% of schools.

Conclusions:

• This study presents the first state-level predictive tool for school drinking water lead leaching risk.
• The model offers a valuable approach to proactively identify and mitigate potential lead exposure in schools.
• Further application of this model can optimize the deployment of resources for childhood lead poisoning prevention programs.