Safety Prediction - Incidents and Injuries in the Workplace
We feel it is critical that the safety industry employ the most advanced predictive analytics capabilities available. It is unconscionable that more time, money and resources are spent using predictive models to track customer loyalty, shopping habits and preferences than in preventing injury and death on the job.
How do we predict workplace injuries before they occur?
Our predictive models draw real-time conclusions about our customers’ future risk using their safety observation data. For instance, our “Red Flag” model identifies projects, sites or work groups that exhibit characteristics suggesting they are at a heightened risk of having a safety incident. Predictive models are extremely useful when the level of data or the number of projects, sites or work groups becomes too large for humans to process.
Our models predict injuries, so our customers can prevent them.
How do we create our safety prediction models?
Our proprietary predictive models were developed in collaboration with world-renowned expert in data learning algorithms, Jaime Carbonnell, Ph.D, and his team at Carnegie Mellon University in Pittsburgh, Pennsylvania.
We “train” models by feeding them actual workplace safety observations collected by our customers, as well as historic work site incidents such as injuries and fatalities. The models use advanced learning algorithms including support vector machines (SVMs) and/or decision trees to learn the relationship between safety observation data and historic work site accidents. Based on these algorithms, the models can identify worksites that have a high risk of incidents. Our available dataset includes more than 100 million observations from more than 15,000 worksites.
How do we know they work?
Our models are rigorously back tested to verify their accuracy. They are also continuously retrained and validated with new data to keep them up to date. In a recent study, the model accurately predicted injuries 84 percent of the time using at-risk behaviors and conditions collected during safety observations and inspections. As technology advances, newer models are expected to improve upon this rating.