Developing a Fatigue Detection Model for Hospital Nurses Using HRV Measures and Machine Learning

Abstract

Fatigue among hospital nurses, resulting from demanding workloads and irregular shift schedules, presents significant risks to both healthcare workers and patient safety. This study developed a fatigue detection model using heart-rate variability (HRV) and investigated its relationship with the Swedish Occupational Fatigue Inventory (SOFI) among nurses. Sixty nurses from a hospital in Depok, Indonesia, participated with HRV data collected via Polar H10 monitors before and after shifts alongside SOFI questionnaires. A mixed ANOVA revealed no significant between-subjects differences in HRV across morning, afternoon, and night shifts. However, within-subjects analyses showed pronounced parasympathetic rebound (elevated Mean RR) and sympathetic withdrawal (reduced Mean HR) post-shift, particularly after afternoon and night shifts, contrasting with stable profiles in morning shifts. Correlation analysis showed significant associations between SOFI dimensions, specifically lack of motivation and sleepiness, with HRV measures, indicating autonomic dysfunction and elevated stress levels. Several machine-learning classifiers were used to develop a fatigue detection model and compare their accuracy. The Fine Gaussian Support Vector Machine (SVM) model achieved the highest performance with 81.48% accuracy and an 81% F1 score, outperforming other models. These findings suggest that HRV-based fatigue detection integrated with machine learning provides a promising approach for continuous nurse fatigue monitoring.