New York City has made its fire risk inspections about 20% more accurate by using a regression analysis tool to determine which buildings are most vulnerable. The tool assesses more than 7,500 risk factors – including a building’s tax status, construction materials and location – using data from 17 agencies and calls to municipal services. The weight assigned to each variable varies in relation to the building and its district’s historic fire patterns. Every day, city fire crews receive a list of the top 15 at-risk buildings in their district.
Results & Impact
Within a month of the first Firecast model being deployed, the number of violations issued across the city increased by nearly 20%. Inspections can never eliminate the risk of fire because they are primarily caused by human error, but as many as a quarter of buildings that suffer fires will now have been inspected within the previous three months. It is thought that this will save civilian and firefighter lives, as fire crews will be more familiar with buildings they enter.
Fire Department of New York City, Department of Buildings, Department of Health, Department of Finance, Department of Environmental Protection
A predictive modelling tool, Firecast, analyses up to 7,500 risk factors to calculate a risk score for each of the 330,000 buildings the New York Fire Department inspects. Factors include previous fire, injuries, the number of buildings close by, where in the city it is, as well as any tax or health violations. Building complaints filed by 311 (municipal services) calls are also incorporated into the model. The tool uses regression analysis to identify correlations between the data and past fires. Factors are weighed according to historic patterns in different parts of the city - meaning that reports of a trash violation in a building in Brooklyn could affect its fire risk score differently than the same violation linked to a building in Queens. The model is run each night and provides firefighters with a list of the 15 buildings most at risk in their district. Data is pulled from a central repository connected to the databases of five city agencies, as well as from DataBridge, New York City's data sharing platform.
New York City
Cost & Value
Running since 2013
Because fires are frequently a product of human error, there is a limit to the degree to which inspections can prevent them. Instead, the Risk-Based Inspection System is primarily designed to reduce the risk posed to fire crews and civilians. Fire crews will be more familiar with the structures in which they are most likely to have to tackle blazes, helping them navigate them quickly, and evacuation routes are less likely to be blocked. In terms of the model itself, using data from across city agencies was problematic because the same information was categorised in different ways. For instance, the Department of Buildings gave each structure a Unique Identification Number, while the Department of Finance tracked buildings using a property tax lot. As a result, when the Fire Department first received data from other agencies, it required extensive cleaning to create a common identifier for all buildings.
New York City’s fire risk inspections are up to 20% more accurate thanks to a tool that predicts which of the city’s buildings are most vulnerable.
The Firecast prediction tool analyses up to 7,500 risk factors (such as a building’s location, tax records and construction materials) using data from across city agencies. Fire crews receive lists of the 15 buildings most likely to catch fire within their district, with factors weighed according to their significance in local areas.
The system is saving the lives of firefighters and civilians alike. As many as a quarter of the blazes fire crews tackle are now likely to occur in structures inspected within the previous three months, giving fire crews more accurate and up-to-date information about their surroundings. Violations given to building owners are now more extensive and handed out more frequently, reducing the likelihood of escape routes being impeded and limiting the impact of poor maintenance on provoking fires.
“One of the things we found was that before the model was deployed, the types of buildings fire crews would go and inspect did not match fire activity,” said Jeffrey Roth, former Assistant Commissioner for Management Initiatives at the New York City Fire Department.
Prior to the Risk-Based Inspection System, structures’ fire risk was measured using a flat, alphabetical grading system between A-E, making it hard to differentiate between the relative risk to different buildings.
This lack of information about at-risk buildings contributed to the deaths of two fire fighters in the Deutsche Bank Building in 2007. It was later discovered that the building had not been inspected for five months, even though it should have been checked every 15 days.
Following the blaze, New York set out to create a system to predict which buildings were most at risk of fire. Overseen by a purpose-built analytics unit inside the Fire Department, focus groups with firefighters were used to identify the factors that best relate to fire risk, providing the foundations for a basic predictive modelling tool. This was developed by incorporating data held by other agencies, such as tax liens (restrictions imposed by government for the non-payment of tax), health violations and the financial status of buildings. Such data could provide evidence of neglect, implying a greater risk of fire hazards. Regression analysis was run on these factors using five years worth of data from serious fires, allowing the relative weighting of each variable to be refined.
The result was a sophisticated modelling tool, known as Firecast. It scores each of the 330,000 buildings the Fire Department is liable to inspect between 0-100, providing a list of the 15 most at-risk structures in each district. First rolled out in 2013, the newest version, developed in 2014, incorporates information from 17 different agencies. The tool also uses information from 311 (municipal services) calls, the majority of which are complaints about buildings.
Unlike its earlier edition, the current tool weighs factors differently according to local circumstances. For instance, if fires in one part of the city are more closely correlated with trash being improperly stored, reports of a trash violation in a particular building in that part of the system will increase its risk score by a greater amount.
“Geography really mattered in this model,” said Roth. “A building in one neighbourhood that is say, wealthier, could have similar characteristics to a building in a neighbourhood that is not as wealthy, yet presents a different level of risk.”
Firecast is linked to a portal which fire crews use to manage inspections. Once they log in, they are presented with two lists of 15 buildings representing the highest priority fire inspection targets. These reflect two different types of inspection: buildings that have to be visited at regular intervals regardless of risk, such as schools, and structures deemed most susceptible to fire. Once a building has been visited, fire crews notify the system that the building has been checked and that any safety failings have been acted upon, prompting the structure to be removed from the list for a set period of time.
“They had some discretion,” said Roth. “They might go to 15 over a couple of weeks or just pick a few that they knew were really bad. It would just point them in the right direction in terms of, ‘These are what we think are riskiest in your district – go look at them’.”
Although the Risk-Based Inspection System is making New York’s buildings safer, it is still not a failsafe way of preventing fires, given the significance of human error in causing them. Nevertheless, the more knowledge firefighters have about buildings in their vicinity, the less likely tragedies such as the Deutsche Bank fire of 2007 are to be repeated.
“By forcing firefighters to go out and inspect buildings where we thought there was a likelihood of fire, there was less chance that there would be egregious things that would inhibit safety,” said Roth. “So when or if there is a fire, people can get out – there are no chains on the door and exit signs are lit. A lot of fires are caused by human behaviour, so those are things that no amount of human inspection can correct for. But if a fire does occur, we can at least ensure that firefighters are familiar with the building and that the public can get to safety.”
New York has between 3,000–4,000 serious fires a year while the city’s 341 fire agencies are responsible for monitoring about 330,000 buildings.
(Picture credit: Pixabay)