A map of New York City that displays the number of serious fire incidents within commercial and high-rise buildings. FDNY uniformed fire personnel inspect commercial and high-rise residential buildings during their regularly scheduled building inspection time.
 
FDNY Analytics Unit
New York City has about a million buildings, and each year 3,000 of them erupt in a major fire. Can officials predict which ones will go up in flames?
The New York City Fire Department thinks it can use data mining to do that. Analysts at the department say that some buildings are linked to characteristics that make them more likely to have a fire than others.
Poverty, for one.
“Low-income neighborhoods are correlated with fires,” said Jeff Chen, the department’s Director of Analytics, at an industry conference in Las Vegas.
Other factors that correlate with deadly fires: the age of the building, whether it has electrical issues, the number and location of sprinklers and the presence of elevators. Buildings that are vacant or unguarded are twice as likely to have a fire, Chen says.
All this may sound obvious. But it is hard to absorb all the relevant factors at once.
So New York officials have taken roughly 60 different factors and built an algorithm that assigns each one of the city’s 330,000 inspectable buildings with a risk score (The Fire Department doesn’t inspect single and two-family homes).
When fire officers go on weekly inspections, the computer spits out a sheet with a list of buildings, ranked by their risk score, that they should visit first.
The data-mining program went into effect in July and will be expanded to 2400 categories in the coming months. Inspections before it was adopted were fairly random, Chen says.
Buildings considered to be safety priorities, like schools and libraries, were supposed to be inspected more frequently. But the city didn’t target specific buildings based on their risk.
What’s happening in New York City–which became more data-driven under former Mayor Michael Bloomberg–is an example of how many municipalities are trying to to use the data they routinely collect to improve services.
Boston uses big data in its Problem Properties program, which exploits information from different city sources–such as complaint calls, safety records, crimes and tax collections–to identify which properties should get visits by police.
While making investments in big data systems seems like common sense, cities have trouble measuring their success. Officials may be able to cite statistics showing the number of fires or crimes dropping, but demonstrating that big data tools were the reason may be difficult because it involves proving a negative–that something didn’t happen because of their efforts.
“Ultimately, we should see the number of fires go down,” says Jeff Roth, the Fire Department’s Assistant Commissioner for Management Initiatives. “And fires should become less severe.”
Reuters