Pittsburgh, PA Fights Fires with Predictive Analysis

Carnegie Mellon University’s $ Metro21$  partnered with the City of Pittsburgh’s Bureau of Fire and Department of Innovation and Performance to deliver a first-of-its-kind model which uses predictive analysis to rate non-residential building fire risk.

Interactive mapping, cleverly named "Burgh’s Eye View,” is helping guide the Bureau of Fire in prioritizing inspections based on risk score with the goal of significantly reducing fire risk, property damage and potential loss of life. Also included is a handy dashboard with allows risk scores to be compared (i.e. by neighborhood, property type, or fire district) and a utility to export the data to a spreadsheet for easy access.

As with most metropolitan areas, capacity limitations prevent Pittsburgh’s fire inspectors from being able to inspect the 20,000+ commercial buildings each year. The use of data to more strategically select the properties deemed high risk is expected to have exponential gains in safety with improved operational efficiencies.

The risk-based model, which was deployed in February of 2018, incorporates historical fire incident data from the Bureau of Fire, inspection data, and commercial property data from various municipal agencies including Allegheny County to predict the likelihood of a fire. Other factors are considered in the model such as high-rise apartment buildings, senior living centers and other predictive features like gas leaks, electrical wiring problems, lot area, etc.

The findings since deployment are promising, with accurate detection of 57% of all (code 100) fire incidents that occurred in any given 6-month window.



Others efforts to reduce fire risk

Very few governments have utilized risk-based modeling to predict and drive mitigation efforts for fire risk.

The New Yor Mayor’s Office of Data Analytics partnered with the Fire Department of New York to develop a proprietary "Risk-Based Inspection System" which was deployed in 2013; however, no information was made available to the public, making is impossible for other local governments to benefit from their good work. We think that’s a real bummer.

In 2015, Altanta’s Fire Rescue Department followed suit with the release of "Firebird" fire risk prediction framework with help from the Data Science for Social Good program. This model was arguably flawed as it incorporated static vs. dynamic data, meaning the model is only able to make predictions based off a snapshot of data. While this groundbreaking model, available as open source is being used as a springboard for other municipal endeavors, such as the $ SF Fire Risk Project$  led by Code for San Francisco's data science brigade and the use of open civic data by Johnathan Jay and Chris Wheelahan to build a $ model to predict residential fire risk in Baton Rouge, LA$ .

Rethinking the model

The Fire Risk Analysis, led by CMU doctoral student Michael Madaio, is the first risk-based model to be deployed in Pittsburgh and the first-of-its-kind in the country.

The model deployed on Pittsburgh Bureau of Fire’s server refreshes weekly with live data from various departments and retrains the model at this same frequency. By leveraging machine learning, Madaio and team have ensured their model is constantly getting better.

Unlike other models which provide observations on census blocks, this model spits out risk score for individual properties, allowing fire inspectors to specifically target higher risk properties, further increasing efficiency and maximizing the impact of fire inspection work.

How other municipalities can leverage this work

The early success of this project has opened the door of possibilities for better use of data and modeling to drive more efficient and strategic operations in local government.

But not all local governments have access to a Michael Madaio (or a local institute like Metro21 eagerly giving a helping hand).

In line with Metro21’s Smart Cities Initiative to “research, develop and deploy 21st century solutions to the challenges facing metro areas”, the open source code for the model is available for use by other cities and the team is working to make this more easily applied to other cities. Also available is their technical report detailing their process, which is available on the Metro21 page and also posted as a linked resource in this story.

Open data opens doors

Local governments are working to various ways to leverage data to improve operations and deliver efficiencies. Pittsburgh’s Department of Innovation and Performance is one of the many local government agencies driving the conversation about more strategic and sophisticated use of data to change the way their city operates. Pittsburgh has worked hard to make data more open and accessible, which made their work with Metro21 on this endeavor much more achievable.

Creating an open data platform is the first step in leveraging data to drive innovative efforts - an initiative many local governments are still working towards. Accessible data enables partners, such as CMU’s Metro21, to help drive government innovation through smarter use of data.

Our community of Ideas

Govlaunch also aims to help municipalities take the next step by sharing their innovative work, the products they use and resources, such as Gitbook pages and Open Data portals to help foster better information sharing and problem solving between local governments of all sizes.

We will continue to follow the innovative work the City of Pittsburgh is doing in partnership with CMU’s Metro 21 and post revisions to the story as the model and project progresses.

A great example of people coming together to get data to work better and work smarter to drive innovation in their community - Beautifully done.



--------

CITATION: Metro21: Smart Cities Initiative (2018). Predictive Modeling of Building Fire Risk: Designing and evaluating predictive models of fire risk to prioritize property fire inspections. Metro21 Research Publication.