Throughout the mining sector, operator fatigue is a leading cause of accidents, safety incidents, and lost productivity. In recent years, best practices for fatigue management have shifted from purely “reactive” methods, such as cameras detecting microsleeps, to a combination of “reactive” and “predictive” methods, where technology predicts fatigue likely to occur later in a shift. Previous work by Fatigue Science has revealed a significant improvement to safety and productivity in the mining industry due to the use of predictive fatigue technology. However, previous iterations of this technology required the deployment of wearable devices (e.g. wrist-worn sleep trackers) in the field, limiting real-world adoption by some mining firms. 

In this presentation, we will evaluate the predictive strength of a Machine Learning model that predicts both fatigue and personal accident risk in mining, without requiring the deployment of wearables. In lieu of sleep data captured from wearables, key inputs to the ML model include an operator’s work schedule, commute, basic demographics, and a sleep questionnaire, which are then analyzed alongside a large training dataset of 5 million anonymized sleeps, which have been recorded by industrial shift workers using wearables on the Readi platform.