PEC Predictive: Innovative, Advanced Machine Learning Incident Prediction
This article is co-authored by Nick Bernini, Manager of Predictive Analytics for Predictive Solutions.
PEC’s latest safety tool, PEC Predictive, is industry’s most powerful tool for predicting the likelihood of an incident.
PEC Safety has partnered with Predictive Solutions, an Industrial Scientific Company, to create an advanced Machine Learning model which predicts the likelihood of a recordable incident over a three-month period. Using leading indicators, rather than lagging data, Owner Clients can manage their finite resources to better manage risk.
Predictive Solutions worked hand-in-hand with PEC leadership to craft our exclusive advanced learning model. The PEC Predictive model scores a Grade A equivalent on the Receiver Operating Characteristics Curve – the highest grade possible.
Each piece of data fed into the model improves it, while also making it more complex. Predictive Solutions’ goal is to build the strongest algorithm, with as little data as necessary. Predictive Solutions spent months combing through actual customer compliance questionnaire data to create the best model possible.
Each learning model is subject to intensive testing against control group data using an industry standard metric called the Receiver Operating Characteristic Curve. This testing measures the model’s ability to separate True Positives (injury occurred) and True Negatives (injury did not occur).
The model’s results are tested against actual data from the field, creating the most accurate model possible. And, as the model degrades over time, Predictive Solutions reconsiders the applicable data to continuously improve the model.
Each month, more than 40 million anonymized contractor data points are run through a proprietary modeling system built by Predictive Solutions’ data scientists. That data is used to rank contractors in three categories: standard; elevated; or high.
Contractors’ ranking is based on Machine Learning technology that identify patterns in the “questionnaire” data, which pinpoints when injuries (risk) are likely to occur. These data points, along with a list of the top three contributing factors, are fed directly back into our Compliance Pro platform for Owner Clients to use when selecting contractors.
Highlighting these risk factors allows clients and contractors to focus their limited risk mitigation resources where they are most likely to prevent a workplace injury.
Every month, all contractor questionnaire data is run through the predictive model and new rankings are assigned. With each iteration of data, the model adapts and the incident likelihood may change slightly. Standard risk means contractors are at an average risk level compared to their peers, with less than 20% likelihood of an incident over 90 days. Elevated represents contractors which have between 21% and 70% likelihood of an incident. Contractors ranked as High have a greater than 70% likelihood of a recordable incident in the next 90 days. This can fluctuate from month to month.
Predictive Solutions understands the needs of the safety industry. For almost 20 years, Predictive Solutions has enabled clients to save lives by predicting and preventing workplace injuries through our industry-leading software solutions, data scientists, and safety experts.
PEC is the only compliance management platform on the market with predictive technology as part of its contractor management toolkit – an innovation that will help create a safer, smarter industry.
Over two decades ago, PEC Safety pioneered standardized learning content and contractor management solutions for industry – revolutionizing safety on a global scale. Now the largest interconnected safety network in the world, PEC’s partners with industry to meet our founding mission: to help bring workers home safe.
With PEC Predictive, a first-of-its-kind innovation, we remain committed to that mission.