Recognizing fatigue using wearable sensors and machine learning

Fatigue has a negative influence on people, it can lead to impaired decision-making and judgement, impaired hand-eye coordination, short term memory problems, poor concentration and reduced motivation In healthcare workers, who are shown to be susceptible to fatigue, this can have consequences for patient safety. Therefor it is beneficial to be able to identify fatigue in the work environment. Wearables provide a non-intrusive option for continuous monitoring of healthcare workers and as relationships have been found between fatigue and walking gait, predicting fatigue based on walking data was seen as a possibility.

This was tested using two accelerometers and gyroscope combination sensors, one placed on the heel of the foot and one placed on the lower back. Walking data was collected of peoples before and after a fatigue inducing activity. These measurements were used as input in a machine learning model which was tested to predict the fatigue state of participants based on their walking characteristics.

For binary fatigue classification, recognizing if people are in a fatigued or non-fatiqued state, the models show to be able to recognize the correct fatigue state, for most people, more than 80 percent of the time. As the results of the models, containing either features from both sensors or just using a single sensor seperately, lay close together, it also shows that there is a possibility of using just one sensor instead of two. This would be beneficial for cost- and use efficiency. Which of the sensors can be used best would require further research and would depend on the requirements of the user, if a higher mean accuracy is desirable or less variance.