Artificial intelligence developed by Carnegie Mellon University can pinpoint which commercial buildings are most likely to catch fire.
The software aggregates risk factors from previous inspection data and fire incidents, and alerts the city's Department of Public Safety when a building is deemed to be high risk.
In the past six months, the software marked 57 buildings--50 of which had some sort of fire event in the following months, such as an electrical fire. There are about 22,000 commercial buildings in Pittsburgh.
Michael Madaio, a doctoral student at Carnegie Mellon University's Human-Computer Interaction Institute, is leading the project.
"We update our model every week with new data to ensure these risk estimates stay up to date," Madaio said. "As Pittsburgh continues to grow and change, we want to make sure our predictive model is able to grow and change as well."
Pittsburgh Fire Chief Darryl Jones said the bureau follows a protocol once a building is labeled at-risk.
"We make sure that systems that are integral to the building are functioning properly," Jones said. "Sprinkler systems, standpipe systems, alarm systems, and we make sure that the residents of those buildings are educated and understand fire safety."
Jones said the most worrisome tend to be high-rise apartment buildings, each with several dozen units harboring lots of people in close quarters.
The CMU team behind the AI is working to expand the technology to predict fire risk for residential properties. They believe their system is the first of its kind.