Newsroom
Machine-learning approach produced a 60% increase in the number of high-risk facilities identified
RESEARCH TRIANGLE PARK, N.C. — A new study led by experts at RTI International, a nonprofit research institute, has found that machine-learning models can be more effective than traditional methods at identifying child care facilities at high risk of having tap water containing lead.
The research team tasked machine-learning models with predicting building-wide water lead risk in more than 4,000 child care facilities in North Carolina. The models outperformed factors currently used to determine risk, which include building age, water source and Head Start program status, by 118–213% based on precision and sensitivity.
“Lead testing programs in the U.S. have long been in need of improved methods for identifying high-risk facilities,” said Riley Mulhern, Ph.D., a research environmental engineer at RTI and lead author of the study. “Our findings suggests that machine-learning approaches could be a significant improvement on current methods, while optimizing resources and ultimately protecting the health of more children.”
Overall, the machine-learning approach produced a 60% increase in the number of high-risk facilities that could be identified and up to a 49% decrease in the number of samples that would need to be collected. The models were most successful predicting facilities where at least one drinking or cooking tap exceeded 1 part per billion, which is the American Academy of Pediatrics’ reference level for lead in children’s water.
To train models to identify and prioritize child care facilities for water lead testing, the research team used data from the Clean Water for Carolina Kids program, the largest peer-reviewed data set of lead concentrations in child care facilities in the U.S.
“The study also confirms previous findings that child care facilities serving low-income families exhibit greater water lead risk at every target concentration, which is a major environmental justice concern,” said Jennifer Hoponick Redmon, MSES, MPA, CHMM, the director of the environmental health and water quality program at RTI.
The study was funded by a federal Water Infrastructure Improvements for the Nation (WIIN) grant provided to the North Carolina Department of Health and Human Services. It was published in Environmental Science & Technology.
