Skip to content

An investingation into London FIre Brigade's callout data.

Notifications You must be signed in to change notification settings

tj2904/lfb-callout-analysis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

London Fire Brigade (LFB) Callout Analysis

Business Problems:

This dataset will be employed to answer the following questions:

  • What are the trends of callouts?
    • By understanding trends resource allocation and be predicted and planned for.
  • How frequently are incidents attended by services from outside the borough?
    • Knowing this will feed into the next question and illuminate how well covered the borough is, with the hypothesis being if it is well covered/resourced within the borough outside help won’t be required often.
  • Are there incident location hotspots within the borough?
    • Along with the previous question knowing this can influence station location and when paired with response time information support the service to hit its response time targets.
  • What factors predict a false alarm?
    • Being able to predict a false alarm could influence resource allocation decisions, every false alarm not attended represents a saving. Of course, false negatives carry potential heavy consequences.
  • What factors are most associated with fires within dwellings?
    • Identifying the key risk factors for residential fires will support the LFB in its target to support the community and where fire prevention work would be best targeted.

Data Pre-processing

Data Types

To ensure that the correct datatypes were allocated to each variable Easting_rounded, Northing_rounded and HourOfCall needed converting to categorical types.

Null/Missing Values

SpecialServiceType null values are not errors but refer to incidents not classed as a Special Service, FirstPump_Arriving_AttendanceTime has 513 null values. The vast majority of these refer to cases of Lift release, all are Special Service calls or False alarms, the FirstPump_Arriving_DeployedFromStation is also missing. Replacing the latter would not be helpful at all. It is inferred that a missing pump time indicates that the incident was attended by staff without a pump, so the missing values will be accepted, replacement would negatively impact the quality of the dataset, but they will be dropped/exclude if used for averaging etc. as to not drag the mean values down.

Imbalanced classes

There is class imbalance within the variables that will be used for both decision tree and k-nearest neighbour (KNN) model construction. This will be addressed using the SMOTE algorithm to over-sample the minority classes. Analysis also highlighted the imbalance in the data intended for use for Association Analysis, however, this is a categorical variable so will need to be encoded before use, this encoding will remove the need to scale or normalise the data.

Model Construction

What are the trends of callouts?

image image

More plots are in the notebook

It is clear that there is a rise in demand placed upon LFB, with more incidents being responded to, with a longer peak period than previously experienced. This has implications for future planning that should be considered.

How frequently are incidents attended by services from outside the borough?

image image

image

More plots are in the notebook

For Hounslow there are a significant proportion of incidents that are supported by response from stations that are not inside the borough. This would benefit from further investigation by LFB, especially to determine if this pattern is unusual compared to the rest of London, and what impact it is having upon the cost of rendering this service.

Are there incident location hotspots within the borough?

image image image

More plots are in the notebook

Key locations within the borough have been identified, and it has been found that stations are located close to these areas. Further investigation could be completed to see how frequently these areas are attended by non-borough stations to identify if current provision is cost effective.

What factors predict a false alarm?

image

More plots are in the notebook

It is clear that false alarms are a key demand upon the resources of LFB, which suggests that any reduction in attending these would yield cost and service quality benefits. Models have demonstrated some ability to predict false alarms, but further work is needed to achieve the levels of accuracy that would be found acceptable.

What factors are most associated with fires within dwellings?

It can be seen from the analysis that PropertyType is a factor that is associated with Fires within dwellings. Specifically, A property type of House – Single Occupancy, or Purpose Built Flats/Maisonettes are the most associated types. This is in-line with what intuition may suggest. It is also indicated that residential fires are more often associated with a low number of calls. What isn’t associated with Dwelling Fires is also, and potentially more so, interesting. For example, large multi-person dwellings such as tower blocks do not appear to be more likely to be the location of a dwelling fire.

Summary

This section presents a summary of the findings along with recommendations for LFB. It is suggested that LFB look into the false alarm rate at Chiswick as this is particularly high, proportionally, for the borough they also attend far fewer incidents, addressing this may support and improvement in the service. The increase in the four wards maybe due to a rise in housing, identifying if this is the case, and reviewing existing plans for further housing expansion would support future service provision planning.