Building Smarter Cities, One Semester at a Time

In 2019, the national estimates remain jarring: Three out of every five fire-related deaths in homes occur in properties without working smoke alarms.

In Louisville, Kentucky, like many cities, public safety officials have long tried to counter such realities with preventative measures—including through public education and smoke alarm distribution efforts.

But how are governments to know where smoke alarms might be needed most? How can they predict which areas of their community might be most at risk of fires, well before the smoke appears?

For years, the solution in Louisville—a city with some 620,000 residents—was one based on intuition and residents’ own requests.

“The way they had been doing that is they would go to community centers, maybe a little bit of door to door, host some events in different neighborhoods,” said Michael Schnuerle, the city’s data officer. “They would think there might be some need (in a certain area).”

A Penn partnership has now refocused those efforts.

For the first time, the Louisville Fire Department is targeting its smoke alarm distribution initiative, and a similar effort relating to vacant properties, based on fire “risk scores” developed by students in the capstone course of the Weitzman School’s Master of Urban Spatial Analytics program.

Those scores—calculated through machine-learning built on data about past fires, neighborhoods and environments—are the latest example of MUSA students partnering with cities to rethink how and where to leverage certain resources, bolstering public policy with evidence.

MUSA’s aptly named Smart Cities Practicum began in Spring 2018 and has quickly logged an impressive list of work. Among the dozen completed projects in the practicum, students have conceived models that predict lead presence in homes in homes in Minneapolis, Minnesota; identify homes most at risk of building code violations in Syracuse, New York; and predict the spatial risk of opioid overdose events in Providence, Rhode Island.

Typically, MUSA Program Director Ken Steif and Weitzman Lecturer Michael Fichman spend the fall semester working with partner agencies to identify potential issues that could be better addressed with data—and then identifying the types of relevant data that exists—to create scopes of work for the students to tackle in the spring.

“My role is to listen to the governments to figure out what their big operational decision-making problems are,” Steif said. “And then think about how this new technology can reduce the friction in solving some of those problems.”

Louisville, which appears to have fully embraced the role of data in recent years, has become one of the practicum’s mainstay clients.

Schnuerle was named the city’s first data officer three years ago and said the partnership—free for the government-clients—has helped fill in critical gaps that most local governments face: “Tools, capacity, skills and costs.”

“It becomes a kind of partnership where we can get things done that we otherwise wouldn’t be able to get done,” he added.

The students assigned to the smoke alarm outreach initiative divided Louisville into small grids to determine which sections of neighborhoods are most at risk of future fires.

Emily Hu (MUSA’19), who graduated from the program in May, said she and her two partners constructed a portrait of the city by scouring detailed data and information about parcels of land, properties, schools, airports, streets, economic conditions—and, of course, past fire data. Much of the data was drawn from the city’s open data website.

“I really cherished the opportunity to work on a real project with a real client,” said Hu, who now works in Philadelphia as a GIS analyst for a commercial real estate firm.

Among key predictors that contributed to areas receiving higher risk scores were distance to properties that are vacant and that have been cited for interior violations.

Hu said part of the satisfaction was putting a data-driven solution in the hands of public safety officials: The final result was an app that the city’s fire department can use to not only identify areas with high risk scores, but potential venues in those areas to host outreach programs.

The fire department is now using the scores not only for outreach efforts for traditional smoke alarms—it installed more than 900 in 2019—but to prioritize the placement of wireless fire detection devices that Louisville is installing in vacant properties.

“The most important benefit is the hopeful reduction in the injuries and fatalities around fires,” Schnuerle said.

A more effective use of smoke alarms throughout the city could also lead to a reduction in property damage, since firefighters could conceivably respond to fire events at earlier stages, and cut city costs by decreasing the time and resources needed to eliminate the fires, Schnuerle said.

The city plans to assess the impact of the new distribution strategy in about a year.

“I’m really glad that what we could help the fire department, a relatively traditional part of the city, [solve] this problem” through data, Hu said.

For Steif, students developing solutions for specific contexts is key.

“If you go and study data science, these tend to be in engineering or computer science programs (and), not always, but they tend to be agnostic to the use case. These curricula are very focused on the algorithms,” Steif said. “And what we’ve learned through emerging knowledge about algorithmic fairness and related ideas is that ... applying an algorithm to a particular solution has everything to do with the context.”

Fairness in, and transparency into, the algorithms are stressed in the program; they’re important for instilling confidence in city and county governments that the final products are above board.

“There has been a significant … pushback among the use of algorithms. And you're mostly seeing that around private companies that have no incentive nor regulatory reason to disclose the inner workings of their algorithms and what you get likely is a very discriminatory outcome, which is very meaningful if that algorithm is giving you a mortgage, or a credit card, or a health insurance,” Steif said. “We teach students how to open up the black box of these algorithms, how to evaluate algorithms for fairness … how to think about the use of these algorithms as a trade off.”

Steif said in most cases, algorithms come down to a balancing act—“it’s up to the community to decide if we can make an impact on something like child maltreatment, or opiate addiction” and if “the benefits of that algorithm are worth the tradeoff and the costs.”

In Philadelphia, MUSA teams have partnered with Philadelphia Legal Assistance, a nonprofit organization that provides low-income families in Philadelphia with free civil legal aid.

Jonathan Pyle, the organization’s contract performance officer who doubles as the group’s resident data guru, cites a recent project in which students used data to pinpoint properties likely to be renting illegally.

Pyle said that kind of work is important in a city that sees a reported 20,000-plus evictions per year. Pyle sat on the Mayor’s Task Force on Eviction Prevention and Response, whose 2018 report recommended a thorough analysis of illegal rentals and evictions in the city.

“The findings they did come up with are informative,” Pyle said, noting that he shared the team’s work with the city’s Housing Security Working Group—which is working to implement the recommendations of the task force.

The MUSA project eyed factors such as interior and exterior violations, tax delinquency and landlords who live out of state. Among the findings along the way, according to the report, was that “data indicate that evictions and illegal rental violations are more prevalent in areas with higher minority and lower income populations.”

The students ultimately created a tool that could help city inspectors prioritize where to conduct house inspections in order to crack down on illegal rentals—and curb illegal evictions.

In addition to serving cities with data solutions, Pyle said he sees MUSA as a strong training ground for students to take their skills to the public sector: “It further expands the number of people who know about the data, who know how to work with the data and can help make big change as a city as a whole.”

The projects are all also posted online through GitHub, a software development platform, making detailed information about the data sources, modeling and code publicly available—and replication achievable. 

The hope, of course, is that other cities follow suit.