Using 10 years worth of photo and project data provided by Boston-based Suffolk Construction, image analysis firm SmartVid, using its artificial intelligence engine, was able to predict roughly one in five safety incidents with 81 percent confidence (four accurate alerts per year, with the fifth being a false alarm) over a three-year period, and was able to predict these incidents a full week ahead of time.
To accomplish this, SmartVid analyzed the images and then fed additional project data (such as project type, weather and project phase) into a multi-layering machine learning predictive model. These preliminary results are seen as encouraging for the growing use of predictive analytics in predicting safety incidents for the architecture, engineering and construction industry.
Interest is growing so rapidly that nine large construction companies have come together to create the Predictive Analytics Strategic Council. The members are Suffolk Construction, which chairs the Council, Barton Malow (Oak Park, Mich.), Clayco (Chicago), DPR Construction (Redwood City, Calif.), JE Dunn (St. Louis), Messer Construction (Cincinnati), Mortenson (Minneapolis), Shawmut Design and Construction (Boston), and Bouygues Group (France). In addition, Aon, a London-based risk and insurance broker, is on board with the effort.
Each council member has pledged to provide 10 years or more worth of their own data in an anonymous format to help build predictive models for safety and risk. The council's mission is to learn and explore opportunities to build new models for areas of risk, and actively discuss operational, social and business impacts along the way. The first phase will be to build on the work started by Suffolk Construction. The second phase will be to understand the operational impact and implications of these predictions.
"The work is not about one company's risk; it's about reducing risk across our industry," said Jit Kee Chin, chief data officer for Suffolk Construction.