Dwelling Fires

Variable Importance

The most important factor for predicting the category of Very Low to Very High risk is the Number of Households with no car/Van.
For most variables, a higher value would suggest a higher level of risk, but this is not always the case; for example, where the percentage of households who own their house is a lower percentage, this may indicate that the risk is higher. Importantly, these are not always linear relationships between the variables and the level of risk.
Most of the top factors are in some way linked to deprivation, which is not surprising, although there are some factors around the built environment; for example, properties with EPC F/G ratings, 1950-75 construction and number of flats.
To target prevention, ideally it will be finding the people/places where these data points overlap.

The model has evaluated 100s of potential risk factors.
These are the top 10 that it identified as giving the most accurate prediction of the risk of dwelling fires within an LSOA.
Short name	Relative importance	Relationship	Origin
Households with no car/van	100.00%	Positive	Census
% Households who own/share own	93.06%	Negative	Census
Income deprivation affecting older people	59.23%	Positive	IMD
Occupancy room rating - fewer rooms than required	56.42%	Positive	Census
People & family household composition fine multi occupancy dwelling	54.61%	Positive	Mosaic
Work transport to work bus tram	49.25%	Positive	Mosaic
Number of households with no adults in employment	48.29%	Positive	Census
% Households - social renting	45.93%	Positive	Census
Number of flats	45.44%	Positive	Census
Households Council Tax Band A	41.69%	Positive	Valuation Office

Using the top ranked risk variables, the model predicts the risk level in each LSOA.
This map summarises the output in terms of whether each LSOA is most likely to be Very Low to Very High risk.