Predictive Analytics for Close Calls in Construction Safety

Extracting knowledge from data on near hits (aka. close calls) might warrant better understanding on the root causes that lead to such incidents and eliminate them early in the risk mitigation process. While a close call is a subtle event where workers are in close proximity to a hazard, its frequency depends–amongst other factors–on poor site layout, a worker’s willingness to take risks, limited safety education, and pure coincidence. While existing predictive analytics research targets change at strategic levels in the hierarchy of organizations, personalized feedback to strengthen an individual worker’s hazard recognition and avoidance skill set is yet missing. This study tackles the bottom of Heinrich’s safety pyramid by providing an in-depth quantitative analysis of close calls. Modern positioning technology records trajectory data, whereas computational algorithms automatically generate previously unavailable details to close call events. The derived information is embedded in simplified geometric information models that users on a construction site can retrieve, easily understand, and adapt in existing preventative hazard recognition and control processes. Results from scientific and field experiments demonstrate that the developed system works successfully under the constraints of currently available positioning technology.