When approached by Google and the University of Michigan-Flint to collaborate on a mobile app advising Flint residents on lead risk levels in their homes, assistant prof. of Engineering Jacob Abernethy and Business prof. Eric Schwartz saw both a critical gap between available data and the city’s recovery efforts — and a way to fix it.
Of the approximately 35,000 occupied structures in Flint, only about 7 to 10 percent are estimated to be affected by lead contamination, largely due to the usage of private water service lines with lead in the pipes which connect individual homes to city-wide water mains.
However, Schwartz said records on service line composition were incomplete, tests were unreliable and no specific pattern of lead contamination could be discerned from maps.
This, both professors said, puts the state’s current effort to replace lead service lines in a precarious situation, because of difficulties in identifying which lines to replace. With limited resources, replacing all the service lines in the city would be unrealistic, and would delay replacements for homes that are actually contaminated, Abernathy said.
“You have 20 million dollars and you can maybe dig up 5,000 homes in Flint,” Abernethy said. “There’s 55,000 (occupied and unoccupied) homes in Flint; which ones do you choose? You can’t answer the question if there’s a service line buried under the ground that costs $500 to dig up and test.”
However, Abernethy and Schwartz’s work aims to solve this issue through statistics. Drawing on a large team of engineering students from the recently formed Michigan Data Science Team, which they advise, they set out to build a predictive statistical model based on existing data to identify which Flint homes’ service lines are most likely contaminated.
“We didn’t really have an idea, but the moment we started seeing the datasets, that’s when it really clicked,” Schwartz said. “That there was a lot of value we could add, and that this was a perfect opportunity for students to get involved and actually make a difference.”
In March and April 2016, the team acquired large sets of construction and environmental data on each individual residential plot in Flint from agencies such as the Michigan Department of Environmental Quality and the Environmental Protection Agency.
Beginning in May, student teams from MDST began to apply machine learning techniques to determine what variables would carry the strongest predictive value for lead contamination.
The students’ findings, while not intuitive, have proven insightful to the service line replacement, Schwartz said. Students found that homes built in certain non-consecutive years would have especially high risks of lead contamination, likely the result of shortages in copper — which is normally used to build service lines — during those years, causing lead to be used as a substitute material in service lines.
Specifically, homes built during the latter years of World War II are at a significantly elevated risk for lead contamination because the war effort caused shortages in copper. A similar risk cluster has also been identified for homes built in the 1920s.
Schwartz also said student work has begun to suggest the location of a home in Flint also serves as a predictor for contamination. While the team is still uncertain on the cause of this correlation, he suggested the location of a home is likely serving as a proxy for another variable that is yet to be identified, such as lead contamination in non-lead pipes caused by seepage from the environment.
By coordinating directly with the Flint Water Interagency Coordinating Committee, Schwartz said his student team has been a factor in identifying and removing approximately 200 contaminated service lines. He added that projects such as these allow students to make significant impacts to society while still on campus.
“As data science and big data are becoming increasingly popular and entering the curriculum, it’s projects like these that really expand the kinds of problems we can solve, the kinds of contributions we can make to business, to organizations, to society,” Schwartz said.
MDST member Arya Farahi, a Physics Ph.D. candidate, said the Flint project has helped transform the student organization from a competition-focused team into a more impact-driven organization.
“It was basically a competition team before getting the Flint data,” Farahi said. “It was the first time we were not doing competition; it was the first time we started doing collaboration.”