Workplace Solutions Safety program management

Safety program management

Can safety incidents be predicted?

Responding is Griffin Schultz, general manager, Predictive Solutions Corp., Oakdale, PA.

Yes they can – and with much higher accuracy rates than other attempts at prediction by the likes of Carnac the Magnificent, Nostradamus and even Punxsutawney Phil.

In fact, a study based on four years of workplace safety data by a team at Carnegie Mellon University showed that workplace injuries can be predicted with accuracy rates as high as 80-97 percent. CMU researchers used a computer employing “machine learning techniques” to make their predictions. That work simply extends the work done by thousands of individual safety managers around the world.

Every day, safety professionals walk their respective worksites and collect data about how safe or unsafe that worksite is. Once the data is collected, they review it and come to conclusions about their future risk. They predict where, when and how their future incidents will occur, and then proactively prevent them. For many experienced and savvy safety professionals, this works pretty well. At some point, though, humans are constrained inanalyzing data and predicting future outcomes.

These constraints are based on the fact that safety managers are human. They are biologically limited in their ability to effectively analyze data they collect, and then come to conclusions or predictions about their future risk. They are limited first by what they’ve learned across the course of their career. A human’s ability to predict risk is limited in part by everything they’ve ever read, seen or heard related to worker safety. Their knowledge base, compared to all the information available in the world, is fairly limited.

Second, humans are limited in their ability to analyze big data sets with many variables. The brain cannot effectively process these types of data sets – they are simply too complex.

Computers, on the other hand, don’t have these limitations. A computer can “learn” by processing every piece of information we present to it – in theory, all of the information available in the world. It can process any type of data we feed it – the information in every safety inspection ever conducted, every incident investigation ever produced and every training manual ever published. Computers also have no problem finding correlations in large and diverse data sets. In study after study, machine-learning systems make better predictions about future outcomes than humans. Although these machines go through the same process to make predictions about the future as humans, they lack some of the biological limitations that humans inherently possess.

Now, this is not a call to replace knowledgeable safety professionals with machine-learning computers. It is a call, however, to augment the safety manager’s risk management tool set with these proven machine-learning techniques. After all, once a computer makes an accurate prediction, preventive measures can’t be taken by machines or computers. This takes knowledgeable safety professionals with the experience to work with their colleagues to improve safety environments. Computers don’t replace safety professionals; they just make their job of analysis easier so they can focus their expertise on prevention.

Machine-learning computers such as Deep Blue, Watson and the systems that run the predictive algorithms behind Google, Amazon and Netflix have proven their effectiveness in other industries and business functions. It’s time we seriously consider their application to safety, and improve our ability to send every employee home safe every day.

Editor's note: This article represents the independent views of the author and should not be construed as a National Safety Council endorsement.

Post a comment to this article

Safety+Health welcomes comments that promote respectful dialogue. Please stay on topic. Comments that contain personal attacks, profanity or abusive language – or those aggressively promoting products or services – will be removed. We reserve the right to determine which comments violate our comment policy. (Anonymous comments are welcome; merely skip the “name” field in the comment box. An email address is required but will not be included with your comment.)