(Won) Hack-Austin Data Hackathon Exploring Data for Health and Safety

madderle
Dec 11, 2017
3 min read

Updated: Dec 12, 2017

This past weekend (12/8/2017 - 12/10/2017) I participated in a data hackathon put on by the City of Austin Office of Performance Management. It was an amazing experience! I got a chance to meet and work with brilliant, passionate people wanting to use data to drive improvements for our city.

Office of performance management has established a strategic initiative with the goal of:

"..working toward the long-term Imagine Austin vision and our aspiration of being one of the most unique, thriving, livable cities in the country, this City Council has chosen to pursue the following strategic outcomes at this time."

My team worked on the safety challenge. We wanted to help the city use data modeling to create hyper focused strategies that drive operational improvements. Specifically we developed:

A model to identify seasonality and trends; and forecast the call volume and average response time for each Response area.
A list of features most associated with late response times.

You can find a link to our Github repo.

Results

My team won the hackathon for the area of Safety, got a city proclamation and going to present to the city council next year (2018)!

Predicting Call Volume and Average Response Times

To generate the Call Volume and Average Response time forecasts I used time series analysis (ARIMA). For the hackathon I created models for the entire dataset and 8/254 Response areas.

This plot shows the total emergency call volume combined with the time series forecast to end of 2019.

This plot was generated from an ETS analysis and showing that there is seasonality in the data. Specifically there is a drop in calls in the first quarter of the year, followed by a spike, then slowly trends downwards.

This plot shows the data and the prediction for the average response times out until 2019.

Predictive Mapping

Additionally, we used ArcGIS from ESRI to visualize the data. Here is what we developed for the hackathon: Where is the fire?

Insights on Response Times

For the hackathon we also used a Random Forest model to determine the best features associated with late responses. To be clear, a late response is a response that exceeds 8 minutes. We found some interesting insights. Namely, most of the late responses were on Mondays around 4pm. This can mostly be attributed to Austin traffic.

We also looked at the problem type; and small grass fires were the most common type of problem when firefighters had longer response times. These small brush fires can mostly be attributed to spent cigarettes and citizens not being able to give a specific address since its often reported on a highway.

Additionally, our model returns areas as well like Travis Country as a problem with late response times.

Using this data, the city can create a targeted campaign for Travis Country around prevented bush fires and properly disposing of cigarettes.

Impact

Other Departments: This kind of data modeling can be used by other departments that care about call volumes and response times (EMS and Police).

Reactive to Proactive: Often times the emphasis is on "response". This leads to people being reactive. Through data and modeling, we can shift to being more proactive and creating hyper focused strategies to attack the root of the issues rather than broad based strategies.

Effective Use of Resources: The absolute guaranteed way to have the best response times is to place a $10 million dollar fire station in every neighborhood. But that is not possible or practical. Departments and cities have limited resources. Data can be used to help answer the question: are we using our resources effectively? Are we sending the right equipment to a response? Are we using the most efficient routes?

Due Diligence: Data provides clarity. The city planners can use data and then articulate to the council that they have exhausted every possible avenue to reducing the problem and its still projected to remain a problem. They can clearly show they have done their due diligence before requesting a new $10 million station being built.