Paper
4F.004 Predictive Intelligence to Prevent Workplace Injury
Published Mar 1, 2021 · Jaimie McGlashan, S. Norris, Steven Armstrong
Abstracts
0
Citations
0
Influential Citations
Abstract
Preventing workplace injury is critical, however to effectively target preventative activities we need an understanding of the future risk of a workplace. Innovative methods from predictive analytics offer an opportunity to predict future risk of workplace injuries and strategically target preventative regulatory activity. Predictive models were built to predict the likelihood of a workplace injury, as well as the occurrence of eight distinct hazard types; mental, body stressing, chemical, vehicle, hit by moving object, hit object with body, sound, and fall injuries. Gradient boosting machine algorithms from Machine Learning were utilised, leveraging a range of administrative data from WorkSafe Victoria, such as past injuries, inspections, incidents and workplace details. The model development process involved collaboration with health and safety stakeholders and subject matter experts. The models varied in predictive accuracy from 69% to 91%, with body stressing injuries having the strongest predictive accuracy. The predictive power of input features offers insight into lead indicators of workplace injury. While there was variation of feature importance across models, features such as past claims, workplace remuneration and geographic location were consistent lead indicators. Emerging techniques from predictive analytics can provide an important evidence base on which to direct preventive approaches. Workplace risk scores produced by the models can inform the implementation of strategic workplace inspections and other initiatives to create safer workplaces. Future model development will involve expanding the input features and outcomes to enhance the utility of this new application of predictive analytics.
Full text analysis coming soon...