Project 2: Fire Incident Research
Fires can cause damage to property and completely destroy things we love. Worst of all it can badly hurt people and cause loss of life. For this research we will be looking at fire incidents in the city of Toronto to classify them as a Major Fire Incident or a Minor Fire Incident to help for future reference and help people learn from them.
The Data
The data is from Kaggle and is called Fire Incidents and was created by Rihanna Namdari. This data is from the Toronto Fire Service which provides fire protection, technical rescue services, hazardous materials response, and first responder emergency medical assistance in Toronto, Ontario, Canada. The dataset however only includes fire incidents declared by the Fire Marshall of Ontario. This data reflects the time period between 2011 to 2018. I will only be looking at data from 2018.
Pre - Processing the Data
For pre-processing, I determined from looking at the data set that there were some columns that were not relevant for the classification process. Here is the full list of columns below.
I decided to drop the following columns; Latitude, Longitude, Incident Ward, Smoke Alarm at Fire Origin Alarm Failure, Incident Station Area, Smoke Alarm at Fire Origin Alarm Type, Fire Alarm System Impact on Evacuation, Fire Alarm System Operation, Fire Alarm System Presence, and Fire Under Control Time.
We went from 24 columns to 14 columns after deleting the ones we did not need. That is all for the pre-proccessing step.
Data Understanding/Visualization
To classify each fire as a major fire incident or minor fire incident we will use Civilian Casualties, any fire with 1 or greater will be in consideration as a major fire incident. Anything with zero Civilian Casualties will be classified as a minor fire incident.
The graph above shows Civilian Casualties. As you can most incidents have 0 civilian casualties however there are still a lot with 1 or more civilian casualties.
Modeling
I decide to use a Decision Tree because it could show the reason why there were civilian casualties or what factors could lead to civilian casualties.
Below you can see the code that I used.
First, we changed the type of data from strings to integers then we split the data from there we trained our model.
Evaluation
From the Decision Tree model, we can see that if a person was rescued by the fire department that would be an indicator of a civilian casualty and the incident would be categorized as a major incident. It is also interesting to note that if a person was wearing something that caught on fire they would be more likely to be rescued by the fire department.
Storytelling
The data tells us that most fires were not major under the rule that no civilian casualties were recorded for that particular fire. According to the data 1200 fires were not minor fire incidents and less than 200 were major fire incidents. The main indicator of a fire with casualties is when the fire department rescued civilians.
Impact
Using the data we can see that most fires end up with no civilian casualties which is the best news you can get out of any fire incident. However, there are still some civilians that get hurt mainly due to fires being started on clothing, which leads to the fire department rescuing them. Learning from this I think it's important for the Toronto Fire Department to increase their community outreach and help people learn how to prevent fires and how to survive a fire.