How communities around the country are utilizing data analytics to
predict a host of risk factors and reduce fires.
BY JESSE ROMAN
FOR YEARS, THOUSANDS OF BUILDINGS in Atlanta were like zombies in a bad B movie: they were out there somewhere, but nobody was sure where or how many—at least not the Atlanta Fire and Rescue Department (AFRD). The department had grappled with the problem for years, a known unknown that created a palpable unease like a pile of unopened hospital bills.
The problem was that no formal system existed in Atlanta to notify the fire department when a new building or business came online. Unlike in other cities, the department wasn’t required to sign a certificate of occupancy. A sometimes frosty relationship with the city building department and a lack of motivation within the fire department to address the problem didn’t help, said Matt Hinds-Aldrich, a member of the Assessment & Planning section of AFRD.
, Las Vegas, June 13-16, 2016
Community Risk Analysis: Using Inspection Data to Develop Programs that Mitigate Risks
Tuesday, June 14, 11:00 AM to 12:00 PM
Ronald Farr, Underwriter Labratories; Morgana Yahnke, University of California Davis
Smart Enforcement Using Data to Inform Fire Inspection and Enforcement Programs
Tuesday, June 14, 3:30 PM to 4:30 PM
Kathleen Almand, Fire Protection Research Foundation; Len Garis, City of Surrey Fire Department; Robert Kinniburgh, City of Charlotte; Roger Parker, Avondale Fire & Medical Department, Arizona
The National Fire Operations Reporting System (N-FORS), Intelligent Fire Data for Today’s Informed Fire Service Leaders
Tuesday, June 14, 3:30 PM to 5:00 PM
Lori Moore-Merrell, International Association of Fire Fighters
Predictive Community Risk Reduction—Using Data Science to Reduce Fires
Wednesday, June 15, 8:00 AM to 9:00 AM
Matt Hinds-Aldrich, Atlanta Fire Rescue Department; Bistra Dilkina, Georgia Tech; Mike Flowers, Enigma.io; Bill Schneider, Experian Marketing Services; Nathanial Lin, NFPA
Community Risk Reduction Shaping the Department’s Direction and Public Education Programs
Wednesday, June 15, 9:30 AM to 11:00 AM
Thomas DiBernardo and Stephen Grasso, Sunrise Fire Rescue; Derrick Phillips, St. Louis Fire Department; Derrick Sawyer, Philadelphia Fire Department
Feature Presentation—Smart Data for Smarter Fire Protection
Wednesday, June 15, 9:30 AM to 11:00 AM
Bart van Leeuwen, firefighter, data expert, and owner of netage.ni
SPECIAL CONFERENCE FEATURE:
The NFPA Data Analytics Sandbox
Curious about the tools that exist to help you harness the enormous power of data analytics? Check out the NFPA Data Analytics Sandbox, which will be featured on the Expo floor throughout the conference in Las Vegas. For more information visit the 2016 NFPA Conference & Expo website.
“The only time we learned of a new building was when one of our inspectors was driving down the road and would notice a building and say, ‘Hey, that looks like something I should inspect,’” he said. “Our best guess was we were only inspecting 10 percent of commercial properties in the city. We just weren’t aware they were there.”
And until very recently nobody had a clue how to solve the problem.
Across the country, fire departments big and small are facing similarly thorny issues with no obvious fixes. Many, including Atlanta, have turned to data analytics.
Hinds-Aldrich will show how Atlanta is using data to address its building inspection problem at the NFPA Conference & Expo in June in Las Vegas. The education session, called “Predictive Community Risk Reduction—Using Data Science to Reduce Fires,” will also include data scientists, academics, and others who will each detail public-safety data-analytics projects that have already been deployed with transformative results. They range from a data project in New Orleans that is helping fire departments target where to distribute fire alarms to a complex computer algorithm developed by marketing firm Experian to help firefighters with outreach and education initiatives.
“I am hoping to show attendees the multitude of methods and approaches people are taking to apply data to the fire service,” Hinds-Aldrich said. “Some initiatives come from government, some from the private sector, some from academia—we are leveraging a lot of ideas and technology that were designed by people well outside the fire service, but who are bringing interesting insights to the fire problem. And it’s really starting to take off.”
Big data’s big picture
Data’s potential to solve complex problems will be a running theme at this year’s NFPA Conference & Expo. Like many industries before it, analytics is quickly changing the game for public safety agencies. Problems once addressed with educated guesses can now be quantitatively solved. The spray-and-pray method of outreach and prevention has been transformed into a finely targeted laser beam.
“It’s amazing the range of programs out there, from very simple spreadsheet-based initiatives tracking inspections to some quite elaborate models that use all kinds of data sources,” said Kathleen Almand, the vice president of research at NFPA. “In a time of reduced resources for fire departments, many see using data as a way of becoming more efficient, and of better using resources.”
In November, Almand and NFPA hosted a “Smart Enforcement Workshop” in Tempe, Arizona, which brought together 15 local fire departments—ranging in size from big departments such as Atlanta and Charlotte, North Carolina, to rural areas like Avondale, Arizona, and Kitsap County, Washington—to present how they are using data to inform fire prevention, inspection, and enforcement.
A goal of the workshop was to learn what was out there, and get these early adopters “to share what they are doing with each other to make their programs more effective and efficient,” said Almand, who will be hosting a panel discussion about the results of the workshop at NFPA’s conference. “Another goal was to see how NFPA might be able to support what they are trying to do and to see if there are things NFPA should be developing in this area to support other jurisdictions.”
Fire departments can use data analytics to identify neighborhoods that would be best served by smoke alarm installation programs and other community fire safety initiatives. Photograph: iStockPhoto
Hinds-Aldrich, who presented Atlanta’s data analytics tool, called Firebird, said that NFPA’s involvement in helping guide the fire service toward data analytics is critical.
“Right now, data and the fire service is like the Wild West, which is a double-edged sword,” he said. “The benefit is that anything you do is new and novel, and completely uncharted territory. On the other hand, you can get taken. It can quickly become a graveyard of silly ideas that never get off the ground.
“NFPA has such a leadership role and I commend NFPA for getting out in front of this whole trend,” he continued. “There is plenty of good and plenty of bad, but I think NFPA can cut through some of the nonsense and fads and fashion and really lead the fire service down this road.”
NFPA has taken concrete steps to position itself at the forefront of the public safety data revolution. Last September, NFPA hired former IBM data scientist Nathaniel Lin as director of data strategy and analytics, and in December, the association acquired a cluster of servers capable of storing and processing huge quantities of data. The servers will house the NFPA Analytics Sandbox, a data depository where vast amounts of public and private data can be shared and analyzed to develop tools to aid in smarter planning and preparedness, inspections and enforcement, resource allocation, and even code development. The first NFPA-developed data project, born from the enthusiasm of the “Smart Enforcement Workshop,” will be a tool to help jurisdictions use local inspection data to develop better and more efficient inspection practices. Currently, NFPA is in the process of gathering all of the disparate data that will inform its model, and is targeting four or five jurisdictions to use as a test case. Eventually, “we want this to be a software where stakeholders can log in through an online portal, enter in their data and our system will score it for them,” Lin said. The plan is to have it completed and available by the end of the year.
“We have a specific concept and we are putting together an advisory panel consisting of folks from the workshop to try to learn best practices,” Almand said, noting that this first project is only the beginning. “This is an ongoing priority for NFPA. There will be other even more sophisticated data projects going forward.”
The case of the missing buildings
Atlanta is a good example of how different organizations can work together to solve big problems using data. To address AFRD’s building and inspection problem, Hinds-Aldrich reached out last spring to Georgia Tech assistant professor Bistra Dilkina, the co-director of Data Science for Social Good. (Dilkina will also speak at the “Community Risk Reduction” panel discussion.) Each summer the program offers paid internships to 10–15 high-achieving college students from across the country to work on tricky social problems using data analytics. For 10 weeks last summer the team set its sights on helping AFRD find which buildings were missing from its inspection database, which of those require fire inspections, and how to prioritize the influx of new inspections.
Dilkina and her students knew that the roughly 2,500 commercial buildings on AFRD’s inspection roles was only a fraction of the commercial properties in the city. To find the missing buildings they gathered various data sets, such as city business licenses, liquor licenses, childcare and preschool databases, census data, even Google listings. They joined the disparate data—essentially formatting it to make it consistent, sortable, and searchable—and found an additional 6,100 commercial properties previously unknown to AFRD that should be inspected based on use and occupancy.
Data analytics can help communities keep track of new buildings and the challenges they may present for building inspectors and emergency responders. Photograph: Shutterstock.
The audit solved one problem, but created several more. AFRD had barely the resources to keep up with its 2,500 annual inspections and now their list had tripled overnight. The department had no idea where to start.
So the Data Science for Social Good team gathered more data, including historical fire data, and returned to their computers. Fifty-eight different parcel-specific variables—fire history, occupancy type, lot size, appraised value, percentage of units leased, for instance—were used to create a risk-based computer model called Firebird. By combing through years of historical fire data and cross-referencing it with the parcel-specific data, the computer identified variables, such as lot size, that are better fire predictors, and weighted them accordingly. Firebird then ran all the numbers and assigned a fire risk score to each property on the fire department’s inspection list, and also created “heat maps” so AFRD inspectors can visually see areas of the city with more at-risk buildings. The program gave Atlanta firefighters a logical roadmap for handling its massive inspection list.
“Now we have spreadsheets of buildings in descending order of risk, and our inspectors are working their way through them,” said Hinds-Aldrich, noting that it’s too early to say what other steps, such as hiring additional inspectors, may be needed. “Their model is done and out there, but it remains to be seen how well it gets integrated or used. It is still a work in progress.”
Seeing what was accomplished in Atlanta has prompted other jurisdictions from across the country to inquire about using their data in the Firebird model. “The beauty of the model is that it’s very transferrable,” Dilkina said. “Different variables could end up being more or less predicative in different cities, but the model would adjust to the data and continue to be predictive.”
The students who worked on the project have decided to make the Firebird model publicly available as an open source code so it can be used and improved by anyone with an Internet connection. This and other projects have convinced Dilkina that public safety is on the verge of a data revolution.
“I think the interest is growing on both the research side and private-sector side” to do this kind of work, she said. “There are more companies looking at this right now, and as the commercial market grows, student interest grows. The two things go hand in hand.”
The Data Science for Social Good program now routinely receives more than 100 applications each summer, from bachelors to PhD students, for the 10–15 spots available.
“I think this generation of students has a very strong desire to use their skills for the social good,” Dilkina said. “All these safety agencies need to do is tap into that.”