WHS Rating

What is this research about?

Using predictive analytics and machine learning to support an evidence-based approach to identifying high risk workplaces?

Timeline of project

This project has been completed. You can learn more about how this work is being applied in the high risk workplaces strategy.

What did the researchers look at?

The research team conducted analysis of various data sets including business demographics, WHS compliance history, workers compensation claims performance and other information sources.

Using a predictive analytics model, this data was considered and tested to identify the predictive power of more than 400 characteristics in determining the risk of a business having an incident in the next 12 months.

This resulted in the creation of a WHS rating, underpinned by the predictive model, which estimates the level of risk of a workplace having an incident.

The accuracy of the WHS rating was tested on historical data proving an 80+% success rate in predicting that a business would have incident, and a 90+% success rate in predicting a business would not have an incident in the following 12 months.

Outcome

This program has delivered a WHS rating for all workplaces in NSW that have an interaction with SafeWork NSW in the last 10 years, that includes:

  • the likelihood of a workplace having an incident in the next 12 months.
  • the risk level of a workplace in comparison to workplaces sharing similar characteristics, be it in term of size, industry, or both.

The researchers worked with SafeWork NSW to integrate the WHS rating into their processes, supporting and enabling other activities that are being undertaken by SafeWork NSW as part of their high risk workplaces strategy.

It is important to note that the WHS rating is an additional source of information for decision makers to consider while carrying out regulatory activities. The WHS rating is not intended to replace the existing decision-making factors used today, but instead to complement and strengthen this intelligence.