Training the brain
A new team of NFPA data experts is developing powerful new tools that tap a century's worth of information and may soon reshape how fire and safety professionals do their job
BY JESSE ROMAN
SITTING ON SHELVES, in cabinets, and on microfilm in the basement archives of NFPA’s headquarters near Boston are thousands of documents that collectively tell the story of fire in the United States. Countless narratives, charts, photos, and maps recall the circumstances behind numerous calamities, from conflagrations that destroyed entire cities to detailed histories of fire in some of America’s most obscure industries and occupancies.
About 90 minutes away, in a data center in Springfield, Massachusetts, sits the latest addition to NFPA’s data depository. Beneath the casing of an unremarkable looking metallic box lies an assortment of circuits and chips that NFPA staff has nicknamed Crosby, after NFPA’s first secretary, Everett Crosby, who, more than a century ago, first championed the idea of collecting data to reduce fire loss. While the reams of paper in NFPA’s archives dutifully recall the past, Crosby has the potential to reveal the future, taking what has already happened to unlock what may come.
Over the last year, Crosby’s spacious hard drive has been populated with about seven terabytes of data obtained from fire departments, code enforcers, the U.S. census, the geographic information systems (GIS), and the National Fire Incident Reporting System (NFIRS), as well as decades of NFPA’s own fire survey records, with more added all the time. Crosby’s Intel Xeon 24 core processor brain hums, scouring the data to find correlations, clusters, and statistically significant variables, all the while reordering and reweighting the data as new information is added. Training that brain is NFPA’s newly hired team of data scientists and analysts, who busily develop the rules and algorithms that turn Crosby’s data from a collection of statistics into tools that could one day reshape how fire professionals do their jobs.
The effort, while technologically sophisticated, seeks to answer a single question, according to NFPA President and CEO Jim Pauley: “How can we take all of that data and put it in a form that helps our stakeholders with their decision making?” The answer, data analytics, has in the last handful of years become more advanced and more accessible than ever before.
Today’s powerful computers are so adept at sifting, sorting, and analyzing the world’s ever-increasing raft of information that the results often seem to bleed into the realm of science fiction. Google’s analytics servers, for instance, have advanced to the point of now being able to predict the timing and severity of the upcoming flu season. When Crosby’s potential is fully realized, the possible applications are no less impressive and also involve feats of forecasting. With enough good data, for instance, analytics could pinpoint which building in a specific neighborhood is most likely catch fire next. It could objectively tell public outreach officers which city blocks to target for the most impactful fire prevention education campaign and even how many lives their effort might save. Data from a firefighter’s career history may even one day help doctors quantitatively determine his or her risk for specific kinds of cancer.
“The potential of data analytics is so amazing, it’s scary,” said Ken Willette, the first responder segment director at NFPA and a former fire chief. He called NFPA’s new data efforts “transformational” for both the organization and the fire service.
“In the past, our reports have been primarily to inform federal decision makers, Congress, and state-level associations, but haven’t always been so useful to individual departments,” he told me. “But data has to be meaningful at the local level. With this system we have the potential to operate at that level and put the powerhouse of analytics behind it to empower local departments. That is a game changer.”
While businesses of all stripes, from Netflix to AT&T, have utilized this magic for years to learn about customers and boost efficiencies, the fire and life safety sector has barely dipped its toes into the analytics pool. NFPA is ready to provide a forceful shove off the edge.
Two years after NFPA’s Board of Directors made knowledge, information, and data focal points of the association’s new strategic plan, the investments are starting to bear fruit. By the end of November, the first NFPA data tool—a dashboard that will allow local departments to map and manipulate their local fire data—will go into beta testing. Twenty other data analytics projects are also in various stages of development, everything from an analysis of high-rise building sensor data to an algorithm that mines NFPA’s advisory services call logs for patterns that could inform training programs, handbooks, and codes and standards.
The most ambitious NFPA data project by far is the development of what is now being called the National Fire Data System (NFDS). As it’s currently envisioned, the NFDS would collect and analyze incident and operational data from thousands of fire departments around the country, enabling each department to harness the collective wisdom and experiences of the group to inform its own decisions and to justify those decisions to policy makers. In September, in a show of faith for NFPA’s emerging data program, the U.S. Fire Administration (USFA) awarded NFPA $1.2 million to develop the NFDS. The International Association of Firefighters, the International Association of Fire Chiefs, and several other prominent organizations also wrote letters in support of the project.
Click to expand
“We are coming into a very exciting time,” said Nathaniel Lin, a former IBM data scientist who was hired by NFPA in September 2015 to lead its data analytics effort. “At this point, NFPA’s data analytics foundation is still being laid, but the building is also starting to appear from the ground.”
Digging a little deeper
Lin, who wears black-rimmed glasses and has a confident professorial air, speaks with boundless enthusiasm about the prospects of bringing the fire protection sector into the modern data age. Analytics applications, he told me, are most potent when the computers are fed data in huge volumes, ideally from a variety of sources. Here, fire protection has a big advantage over many industries. Most businesses have only their own data to analyze; sharing information among competitors is sacrilege. Fire protection, however, has few of those barriers—fire departments, for instance, aren’t usually in competition with each other—and benefits from a uniquely positioned partner in NFPA.
“Most sectors have trade associations, but they are always one-sided—you’d be hard-pressed to find an organization with NFPA’s stature that does not receive any donations and is not beholden to any special-interest groups,” Lin said. “Every sector would like to do something like the NFDS, but one reason they can’t is that nobody has enough universal trust to be the custodian of that data. There isn’t anything like this in banking, in health care, you name it, because the data exists in silos. Not even government can do it.”
Lin described the century’s worth of untapped raw data collected by NFPA and the fire service as “a gold mine.” In 1896, a group of like-minded insurers came together to discuss automatic fire sprinklers, a new invention with the promise to dramatically curb fire losses. The new devices performed unevenly, however, in part because no one had settled on a right way to install them. So the insurers formed a committee to standardize sprinkler installation, and its first order of business was to seek out fire data to inform its work. Several months later, the insurers founded the National Fire Protection Association, and the sprinkler committee went on to craft what is now NFPA 13, Installation of Sprinkler Systems.
As NFPA grew and the scope of its goals expanded, so did the original notion that better fire protection strategies required studying the fire data. By the early 1920s, NFPA was collecting and cataloging detailed information on about 3,500 fires each year. Incidents were sorted by fire types, hazards, and industries and included data on sprinklers, how and where fires started, how they spread, and how they were extinguished. It was such a massive and costly undertaking—comprising between five and 10 percent of the association’s annual expenses, according to a 1923 report by NFPA’s Fire Record Department—that members debated whether it was worthwhile to continue the effort.
“Have we already continued the compilation of these statistics long enough to definitely prove the fundamental characteristics of the fire loss?” then–NFPA Assistant Secretary R. S. Moulton asked members at NFPA’s 27th annual meeting in Chicago in 1923.
The answer was a resounding “no.” On the contrary, argued a man named Henry A. Fiske at the Chicago meeting, NFPA could be doing much more.
“I think very few in this association realize the vast mine of information we have in Boston,” said Fiske, who had previously overseen NFPA’s Fire Record Department for nearly a decade. “We have taken out a little of the pay ore but there is a lot that we have not yet begun to take out, and I think this association can well afford to try and dig a little deeper.”
And deeper they dug; NFPA’s data efforts today are extensive. The association’s Research Division publishes dozens of reports annually, from big-picture studies on aggregated national data to detailed accounts of specific fires and statistics on fire in an array of industries and occupancies and involving myriad hazards. Meanwhile, the NFPA-supported Fire Protection Research Foundation funds and facilitates dozens of new research projects each year that generate data on topics where there previously was none.
Unlike in 1923, no one has questioned the utility of these efforts, but about two years ago changing needs and new technologies led NFPA to another Henry A. Fiske moment. Could the organization “dig a little deeper?”
“Our community is changing, the problems they are facing are changing, and NFPA’s stakeholders are waking up to the power of real-time local data to make decisions,” Kathleen Almand, NFPA’s vice president of research, told me. “We had a terrible fire problem in this country when NFPA got started. Fire loss reports were very helpful then in convincing communities to spend resources to bring the fire loss down, but now we are in a different place. To get the fire problem down to zero, we have to target resources and solutions, and the way to do that is to target granular use of data. That is the sea change here.”
The idea of using data in more novel ways crystallized for Almand in late 2015 when NFPA hosted an event called the “Smart Enforcement Summit” in Tempe, Arizona. NFPA invited 15 jurisdictions that were already using data analytics to drive their decisions to present their strategies. Some used data to map high fire risk areas, others to inform education efforts or to make budget decisions. One presenter, recently retired Fire Marshal Roger Parker of Avondale Fire and Medical in Arizona, showed attendees how his department used data to map the most fire-prone areas and prioritize fire inspection and prevention efforts. Despite his prevention staff being slashed 70 percent, the prevention division has thrived since the data system was adopted five years ago.
“I was in prevention for 30 years, and a firefighter for 15 years before that,” Parker told me. “All that time we were reactive, we were going by the seat of our pants and our gut reactions. This data program totally changed things. It was the first time I was able manage the program rather than it managing me.”
The Smart Enforcement event “was a huge eye opener for us about what is going on out there,” Almand said. Since then, NFPA has continued interactions with several presenters from the meeting, including Parker, who have shared their data and expertise and serve as advisors as NFPA develops data tools for jurisdictions that might not have the resources or ability to develop their own.
Willette, who in his role as first responder segment director travels the country visiting firehouses, told me he believes that, out of about 31,000 fire departments in the U.S., “the percentage using data analytics is in the single digits, maybe five percent.” Though, he added, most are intrigued by the prospect.
When Parker presents Avondale’s data initiatives at fire service meetings and conferences across the U.S., “the biggest feedback I get is ‘I don’t have the time or resources to do all this,’” he said. “They don’t have any idea where to begin, so they just don’t do it.”
At one presentation, a fire marshal stormed out of the room in the middle of Parker’s talk. When Parker caught up with the man later, he told Parker that the presentation frustrated him. “He told me, ‘You are so far ahead of us. The city will never give us resources to do this, and it pisses me off,’” Parker said. “But that’s where NFPA is coming in with the PIP [Property Inspection Prioritization] program to build a tool that the fire service can use without having to spend much, if any, money. As these tools develop, as people are educated … in five years everyone is going to be using this or they are going to be left behind.”
Enormous potential, uncertain outcomes
The power of analytics has moved in step with the power of computing. With the press of a button, a machine can in milliseconds dig through a pile of digital data that would fill the Library of Congress and find meaningful patterns and hidden correlations that a team of researchers could never hope to uncover.
Here’s one example. Plotting a point on two axes is simple; you find the corresponding X and Y values and make a mark. Plotting a point on 100 axes—taking into account all 100 variables in 100 different dimensions—is much more complex, but no problem for Crosby. If and when the National Fire Data System launches and Crosby is filled with data from fire departments across the country, creating such a matrix could be an immensely powerful tool.
Imagine each fire department plotted with a single dot on a virtual area comprising dozens or even hundreds of axes jutting in every direction—each axis representing one statistical variable about each department, such as the number of fire stations it has, the population it serves, and number of fire calls it receives per year. As each department is plotted with a single dot in this multidimensional space, cluster clouds of similar departments take shape. Though departments in each cluster are remarkably similar to one another in many ways, there inevitably are key differences, too. Looking at those differences is where the opportunity exists, Willette told me.
Say, for instance, that the California cities of Long Beach and Sacramento are similar in many ways—demographics, number of firefighters, housing stock—except that Long Beach has 1,000 structure fires per million residents and Sacramento has only 600. Why would that be? Does Sacramento have more fire stations? Does it invest more money in fire prevention education? Does it have laws that mandate sprinkler installation? By analyzing the data in the NFDS, Crosby will be able to give Long Beach fire chiefs the answer: let’s say in this scenario that Sacramento’s higher fire-prevention education spending is the cause of it having fewer fires. Crosby could then cross-reference Long Beach data against hundreds of other similar departments across the nation and predict how many fires Long Beach could prevent with more education spending, as well as how many lives and injuries could be prevented. Further, it could tell the chief the areas in the city where targeted education would have the biggest impact.
“So now, when I go to the city’s funding authority and say I need $12 million to triple our fire prevention education, I can also say that the $12 million investment is forecast to reduce property losses by $80 million, reduce fire-related fatalities by four, and reduce injuries by 1,000,” Willette said. “Here’s the cost, here are the savings, and here’s all the data to back it up.
“Firefighting funding and budgets have been a shell game,” he continued. “It’s hard to deliver direct measurable performance indicators that policy makers and budget people can look at to compare different budgets and different outcomes and what provides the best value. With data analytics, you can answer those questions.”
The power of benchmarking is just one example of NFDS’s potential. If enough data can be effectively captured, analytics could unlock untold numbers of other truths, such as where best to locate fire stations, how long before a fire apparatus or another piece of equipment will fail, or even a firefighter’s future health risks based on past exposures. “Really, the sky is the limit in terms of what we might be able to do,” Almand said.
But how pie-in-the-sky is all of this? How will the data be collected? Will the fire service be willing or able to provide it? It’s still too soon to know the answers. The NFPA leaders I spoke with admitted that the project’s goals are lofty and will be difficult to achieve, but not impossible.
“It’s big, and as people have probed around with me about that particular project, I’ve been very clear to say that I don’t know for a fact the problem can be solved,” said Pauley, NFPA’s president. “But I do know that if somebody is likely to solve it, it’s us. Because of the relationships we have, the respect we have, and our determination about how important that data can be to what the future of fire safety can look like, things stack up well in our favor.”
At the end of the 18-month USFA grant, NFPA will have developed a roadmap for the project that brings the logistics of the NFDS into better focus. Not long after that, NFPA hopes to begin actually building the infrastructure and testing applications.
“We want to show some early-days progress,” Almand told me. “One of the concepts of analytics is ‘act, learn, build.’ You just have to try it to figure it out, test and see what works, and make it better. We aren’t going to wait 10 years and launch some big system.”
The same philosophy permeates the other 20-and-growing data projects currently on NFPA’s docket and for that matter the analytics initiative as a whole, Pauley said. It’s all a work in progress, a series of trials and errors, a dynamic effort that is difficult to define. In a nutshell, it’s an ambitious unknown with enormous potential and uncertain outcomes.
At the conclusion of the data discussion at NFPA’s 1923 annual meeting, members ultimately voted to expand NFPA’s data gathering efforts, forming a three-person committee to look at how the association might be able to obtain previously unknown national fire death statistics. At the time, it was an immensely ambitious task, because death certificates generally listed only the direct cause of death, such as suffocation or falling, not accounting for the fact that the “falling” occurred when the deceased jumped from a two-story window to escape deadly flames.
Just as NFPA leaders in 1923 had no idea then what the fruits of those efforts might bear, neither does Pauley. But it is well worth the effort, one that could very well change how NFPA is viewed in the future, he told me. “It’s not just about codes, it’s about the data that we can supply, and the applications we can provide to assist people in their jobs,” he said. “We are laying that groundwork so that NFPA will be thought of as a resource in this analytics space, whatever form it takes.”
When I asked him to try to pinpoint where NFPA may be on that timeline—how far have we come? How much further to go?—Pauley doesn’t take the bait.
“I get asked often, ‘Ok, so when we’re done what does this look like?,’ but we’re not going to be done,” he said. “Being somewhere implies a beginning and an end. I think we are well into the journey, but our expectation is that that road in front of us is endless.”