Presentation
Multimodal Predictors of Fatigue in a Naturalistic Occupational Environment: A Forecasting Study
DescriptionFatigue remains a perennial issue in the oil and gas extraction (OGE) industry that has led to injuries, loss of life, efficiency decrease, and property loss. This study aimed to develop fatigue forecasting models for an OGE setting using feasible measures and identifying key features contributing to model accuracy. 70 participants were recruited, and various measures were obtained: demographics, self-reports, sleep, performance (Psychomotor Vigilance Task (PVT)), and physiological measures (heart rate variability(HRV)). The measures were then grouped in order of feasibility, starting from a combination of demographics and self-reports, then adding sleep and performance. These measures were then used to predict a three-class composite label made up of performance (PVT lapses) and self-reports (Rate of Perceived Exertion) using various machine learning algorithms. Using all features, the Random Forest (RF) algorithm achieved the highest accuracy of 57.29%, while the Long-Short Term Memory (LSTM) model achieved similar accuracy of 52.54% without HRV data. When the RF model was analyzed, demographics, PVT, and self-report measures were the most impactful contributors to fatigue predictions, highlighting the need for multimodal markers in complex occupational settings.
Event Type
Lecture
TimeWednesday, October 21st11am - 11:20am PDT
Location
Similar Presentations