IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

New York City Fights Fire with Data

Analytics help New York City firefighters track potential hot spots.

Complex, analytics-driven programs have created a whirlwind of “smart” technologies to help modernize manual tasks and make jobs more efficient. In most industries, smart tools have been developed that try to predict the future so decision-makers can better use resources and personnel.

Predictive policing, for example, helps law enforcement pinpoint where a crime is likelier to occur on a given day, using a number of factors about a certain location. The technology can help alter daily beats, hopefully putting officers in the right place at the right time to prevent crimes.

But police aren’t the only emergency responders benefiting from smart technology — it’s helping firefighters in New York City get a jump on battling blazes too.

The New York City Fire Department (FDNY) has been using the Risk-Based Inspection System (RBIS), an Oracle-based program with data-mining capabilities, to better anticipate where fires may spark. The centerpiece of the tool is FireCast, an algorithm that organizes data from five city agencies into approximately 60 risk factors, which are then used to create lists of buildings that are most vulnerable to fire.

Eugene Ditaranto, chief of FDNY’s Battalion 51, started developing the computer-based RBIS and FireCast in 2008 to better handle the rising volume of building inspection requirements. Previously all inspection records were kept on paper, using a card system. But the records were stored at each firehouse, so there was no way to share information among other fire companies, battalions or divisions.

RBIS provides an inspection data warehouse the entire FDNY can access. Instead of being broken up among the city’s 49 fire companies, any unit can now look up the latest information about when buildings and structures were last inspected.

Ditaranto explained that RBIS excels at coordinating inspection activities for the FDNY’s operational units. Each unit performs 26 different inspection types for the approximately 350,000 buildings in the city. Those multiple tasks were all originally tracked on clipboards.

“You can imagine how cumbersome that can become and how difficult it is to manage,” Ditaranto said. “This system digitally coordinates all of that and understands our various business rules.”

The result is a system that automates the FDNY’s entire inspection workflow, collecting statistics from the inspections and helping management supervise inspections and meet internal goals.

System Evolution

FireCast is the risk-scoring engine driving RBIS. In its first iteration launched in 2010, the algorithm was rather rudimentary, said Ditaranto. It only looked at about six items. RBIS is currently powered by FireCast 2.0, which has 10 times the power of its predecessor. Each data element is given a weight to appropriately calculate fire risk.

Inspectors enter data into RBIS once they finish that day’s scheduled inspections. They then input information into the system about the building so RBIS can recalculate the risk score. The system generates new risk profiles daily.

While RBIS runs smoothly and efficiently, development wasn’t easy. Ditaranto said he and the other architects of RBIS and FireCast — Assistant Chief Edward J. Baggott and Battalion Chief Joel C. Gerardi (retired) — faced much red tape in trying to collect data from other agencies. Complicating matters was the fact that each agency had its own classification system for buildings. So a universal identifier was established so all agency systems could speak to one another.

Ditaranto and the development team had to foster a relationship with the city Department of Buildings. He called the department “the most significant player” in building inspections, as the building code largely prescribes a structure’s design requirements.

More meetings took place with the departments of Health, Finance and Environmental Protection, among others. The players agreed to create a central data hub where all city agencies could feed information. RBIS then accesses that hub for the data needed to evaluate fire risk.

The information is where Ryan Zirngibl comes in. As the head data scientist for the FireCast algorithm, he designs, maintains and updates the process by which all those streams are interpreted and the fire risk for each building is calculated.

Although RBIS and FireCast are proprietary, FDNY has presented about its system to fire departments nationwide. Ditaranto noted that while FDNY is one of the world’s largest and busiest fire departments, public safety concerns are the same regardless of jurisdiction. The technology can be scaled to any size and scope.

“RBIS could be tailored to any particular agency,” he said. “It might not require all the business rules NYC has, but certainly the risk algorithm and the theory behind that could be brought to anybody. But it would really take an agency being willing to invest some time in the development.”

Looking Forward

Despite its success in modernizing how fire risk is calculated and building inspections are done, the FDNY isn’t resting on its laurels. FireCast 3.0 is under development and could be a game-changer in the city.

The third-generation algorithm will examine 7,500 factors across 17 city agency data streams. Ditaranto revealed that the new FireCast will also feature an element of artificial intelligence to track trends citywide.

Zirngibl said there are multiple challenges associated with leveraging such a large amount of information. Computational resources and how the FDNY stores that data is at the top of the list. Zirngibl called it the proverbial needle in a haystack in sifting the useful data from the clutter.

Ditaranto envisions a machine that would notice trash violations in the South Bronx, and then if there were a fire in the same building within 90 days, the program would learn and give trash violations in that area a higher risk rate when computing the area’s fire threat level. But the artificial intelligence must also recognize differing timeframes and predict issues based on numerous measuring tools.

In other words, the AI can’t “cry wolf” every time an incident occurs. FireCast 2.0 looks at constant factors about a building. But if successful, its successor will examine behavioral characteristics that could raise the bar on FDNY’s ability to compute a more accurate risk assessment.

“It will be a very sophisticated algorithm when it’s implemented,” Ditaranto said. “We feel it is the right way to go because risk is dynamic and so are the variables. We can’t have one side be static. They both have to be dynamic, and the machine has to be constantly analyzing that.”

FireCast 3.0 was originally slated to be online and running RBIS this year. But Ditaranto said staffing challenges have pushed back the project’s completion and testing schedule.

Zirngibl confirmed that the team working on FireCast 3.0 is short-staffed, which has slowed development. Tasks like simulations to test accuracy aren’t happening as quickly as first planned, for example.

“This type of analytics requires that the analytical team work closely with the IT team in charge of instituting these changes to the current system — so that what is implemented is what was designed and tested,” Zirngibl said. “Without the staff to ensure this oversight, or to vet this model, progress has slowed immensely.”

Ditaranto added, however, that he’s optimistic FireCast 3.0 will be complete by the end of 2015.

Brian Heaton was a writer for Government Technology and Emergency Management magazines from 2011 to mid-2015.