AI Case Study
The London Fire Brigade identifies the areas prone to fires in ducting systems to provide targeted maintenance using natural language processing
The London Fire Brigade has leveraged artificial intelligence from ASI Data Science to combat fires in London. Over 37,000 free text reports collected since 2000 were analysed using natural language processing to discover patterns. The LFB found that ducting fires are widespread in areas like Soho and along Edgware road and as a result they can now make informed policy decisions, more targeted maintenance and fire inspection visits and ultimately reduce fire deaths in London.
Industry
Public And Social Sector
Public Services
Project Overview
"To identify topics that describe fire scenarios within the collection of 37,000 fire reports [they] used a topic modelling approach called Latent Dirichlet Allocation.
The topics extracted are fairly intuitive, describing fires involving for example unattended candles, the disposal of cigarettes, or electrical faults. The most common topic found describes fires in dwellings, with words such as smoke, flat, and floor as well as fires involving an accumulation of grease, fat, and oil in restaurant extraction and ducting systems.
Let's take the latter as an example. [They] found approximately 300 of these ducting fires. In recent years, there has been a significant increase in the number of ducting fires, despite the general decrease in fires in London.
The topic modelling approach along with categorical data collected by the LFB allowed us to explore more subtle fire types that may not be well recorded by the structured data set. [They] created a tool to explore the geographic distribution of topics for ducting fires. In the below image each red circle represents a ducting fire, where the circle radius indicates the fire severity, larger circles are larger fires.
For the first time, [they] have given the LFB the technical expertise to analyse these free text reports and effectively use data to make more informed decisions, guide policy, and thereby help to reduce fire deaths in London.
Reported Results
"It's evident that ducting fires are particularly widespread in areas like Soho and along Edgware road. This is the kind of information that could be used to provide more targeted maintenance and fire inspection visits in these areas."
Technology
Function
Strategy
Strategic Planning
Background
"Fires in London have been in steady decline since the 1980s. In 1987 fires in London killed 190 people, in 2016 the death toll due to fires fell to 36. Data is crucial in reducing fire fatalities even further.
There are three main reasons for the decline in the number of fires in London since the 1980s. First, technologies have improved including fire retardant materials, smoke detectors, and sprinkler systems. Second, the London Fire Brigade’s (LFBs) active education and outreach programme has been very effective at raising awareness of fire safety issues. Finally, data has been methodically collected and used by the LFB to discover patterns, which has been key to identifying fire risk, leading to informed policy and decision making.
The LFB collect a lot of data; each incident attended is recorded in a data set with 76 multiple choice fields, including information about the time, location, and ignition source of a fire. The business intelligence unit at LFB do a fantastic job of analysing these data. In addition, Fire Investigators also write free text reports for more serious incidents. Since the year 2000 the LFB have amassed an archive of over 37,000 free text reports. This data gold mine has taken hundreds of hours to write, yet historical reports have been sitting dormant because it is difficult to quickly extract information from this volume of free text. If a human were to attempt to interpret this it would take 40 days, and 40 nights to read it."
Benefits
Data
37,000 free text reports