Computer models provide a proactive approach to jobsite safety that can translate into reduced injuries and more profits.
To win business, several construction business owners have turned to predictive analytics to drive their safety programs. This is especially true in the current economic environment that has led to fewer bidding opportunities with increased bidders and decreased profit levels on those opportunities—sometimes below a competitor’s cost.
How are Predictive Safety Analytics Defined?
Ph.D. researchers at Carnegie Mellon University have developed safety prediction models that return high accuracy rates.
Predictive safety analytics can take several forms. But at its most basic level, it is nothing more than collecting safety data—generally, safety inspection or audit observations on work-site conditions—and then modeling that data to predict when and where safety incidents will occur.
Once incidents can be predicted, corrective actions can be taken to prevent them. By taking a proactive and preventive approach to safety, companies can clearly differentiate themselves from their competitors and communicate added value to prospective customers.
How Do Predictive Safety Analytics Work?
Predictive safety models are built by giving computers safety inspection or audit observations. The computers then find the patterns in the data and construct a predictive model. Researchers test the accuracy of these models by giving them a different set of observation data and asking the models to predict the number of incidents. These incident predictions are then checked against the actual incidents, which are reserved and not exposed to the models (for accuracy).
Once these models are fully tested and used, contractors can simply feed the model new and current observation data, and it will predict, with accuracy rates of 80 percent or better, what future incident levels will be.
How Can Predictions Be Used to reduce injuries?
Generally, these predictive models allow companies to focus their resources on their highest risk projects, work teams and activities. (Projects not flagged as high risk should still not be ignored.) With a robust inspection or audit safety technology system supporting this model, safety professionals can drill into the data associated with those at-risk projects to prevent the predicted incidents from occurring.
For instance, the model may indicate that one of your 12 projects has a greater chance for safety incidents. If you examine the data associated with that project, you may find 36 at-risk conditions and behaviors associated with lockout/tagout. You can then take several actions. For example, you could have a safety stand-down to review key lockout/tagout safety procedures with the work crew. Or, you may find the at-risk observations center around one team that might need further training. You might also discover the team has been using faulty equipment that needs to be replaced. Regardless of the findings, the model directs your attention to the corrective actions you should take.
The machine-based models can be fed nearly an unlimited supply of data and can analyze the data in a fraction of the time it takes humans to do the same analysis. If you have a large company with many projects, the analytical task can overwhelm even your strongest safety professionals and most intelligent number-crunchers.
What Can Support the Safety Prediction Models?
To execute predictive safety analytics, actual jobsite safety data must be collected through standard safety audits or inspections. The predictive models can reach their accuracy levels with eight weeks of data and can account for the subjectivity and inconsistency of data collected by human beings. The more data collected, the better the predictive modeling will be.
It is also helpful to deploy a robust safety inspection software system to support the predictive model. With this software system, the safety data can be scrutinized after the model has processed it. And the software systems allow for easier and faster data collection through various mediums such as scanners, computers and smartphones.
Predictive models can identify risk areas and corrective actions, but your employees must carry out these corrective actions to reduce the risk. Your team must be ready to act on the information the models produce to improve the safety culture.
