Mapping fire risk

x Sponsored content on Research Live and in Impact magazine is editorially independent.
Find out more about advertising and sponsorship.

London Fire Brigade is the busiest fire and rescue service in the UK, covering 33 boroughs, more than 4m homes and a million businesses.
The work of a fire brigade doesn’t start and finish with emergency response. Aside from responding to fires in London, the service is firmly focused on prevention. It carries out a number of efforts in this area, including around 80,000 home visits and 1,000 school visits a year to give people advice on how to prevent and detect fire.

In the past decade, fires in London have reduced by 34%, with just under 18,000 fires attended in the city in 2019, compared with more than 27,000 in 2010, and LFB has attributed the reduction to this integrated approach to prevention.

Data is key to informing all of these efforts, from interactive dashboards using Transport for London (TfL) road data to GPS data-mapping the routes fire engine drivers take.

LFB is also using data science and analytics to understand more about the risk of fire and improve the understanding of where to target preventions. A large part of this is the ability to combine different datasets, including publicly available information.

Historically, fire risk has been hard to quantify because properties or people who have had fires are not easily distinguished from those who have not. However, there has been an observation that certain demographics are more likely to have a higher number of fires.

Yet, beyond demographic data, there are various other information sources. Every addressable location in the UK, for example, has a unique property reference number that is looked after by an address custodian – and one recent project focused on applying analytics to these rich, open data sources to try to predict risk.

LFB initially partnered with data-science company Faculty, which runs a fellowship programme that allows postgraduate students to work with organisations.

“We wanted to explore a new dataset that we hadn’t used before, like energy performance certificate (EPC) information, which is publicly available on the Ministry of Housing website,” says Apollo Gerolymbos, head of data analytics, insight and reporting at LFB.

However, the EPC dataset didn’t match the unique property reference numbers, so LFB also worked with GeoPlace, which is responsible for
this data. Other sets, including information on building height, information about fires and other emergencies attended by LFB, and mosaic demographic data at a household level, were used to build an all-addresses corporate database (AACD).

Once the data had been cleaned and the unique identifiers matched, the second stage was machine learning. Various models were trialled and LFB historical incident data was used to train the model and test predictions about historical fire risk.

The project found that EPCs were a better predictor of fire risk than other factors such as building height or property type – and demographics.
“At the end of the project, we saw that the EPC was a better predictor of fire risk than demographics; the idea being that the circumstances that someone lives in may influence their behaviour, which may influence fire risk, rather than it just starting with demographics, which is quite a step change in the way we think.”

In future, findings from this type of analysis could be used in operation so that when crews target individual homes for prevention messaging, they use the most effective dataset of predicting fire risk.

New categories 

In another project, neurolinguistic processing (NLP) was used to analyse the reports produced after every serious fire.

Firefighters must complete an incident record that involves selecting the category of fire from various drop-down menus and categorical fields. But for serious fires, the fire investigation team records free text of up to 500 words about the incident, containing much more information and context – typically, information that would not be captured in the categorical fields or drop-down boxes. Using NLP algorithms meant that topics in the text could be identified in some cases, highlighting new categories of fire.

While some topics uncovered were well known and categorised already, such as fires caused by smoking, the exercise showed others not previously recorded.

More recently, LFB has been exploring whether dirty restaurants could be at a higher risk of a fire becoming serious. It ran a hackathon with analysts from different fire services to explore the question. In a similar approach to the EPC project, food-hygiene ratings data was downloaded from the Food Standards Agency to see whether a lower rating could infer a higher risk of fire.

While insights from these projects have not yet been implemented more widely within LFB, they form an important part of a business case to build the data-science capability, explains Gerolymbos, who is in the process of expanding the team. “The understanding at a strategic level has been about the importance of data science and analytics in the future of the fire service. They’re really valuable projects, because they help communicate why we’re asking to grow the data-science function in the first place.”

Another focus for Gerolymbos is trying to understand how resource-intensive different types of fire are – a fire in a rubbish bin requires far fewer resources than a fire in a flat. “Both of those incidents will appear as one fire in the dataset, so you spend a lot of time trying to understand the severity and how resource-intensive a year has been roughly – other than how many incidents we’ve attended – and trying to look at the utilisation of our resources differently.” This could include assessing whether or not resources were tied up and for how long, he adds.

In June 2017, the Grenfell Tower disaster in Kensington cast a shadow over the country, highlighting the devastating consequences of fire. The LFB is awaiting the outcomes and recommendations from the ongoing Grenfell Inquiry, as well as other reviews including Her Majesty’s Inspectorate.

These mean that it is the right time to be asking new questions and exploring and analysing new information, according to Gerolymbos.
“All services in the UK are going through the same process at the moment, but projects like this in London are feeding into research about how things are done across the UK.”

This article was first published in the July 2020 issue of Impact.