Classification of Physical Fatigue on Heart Rate by Wearable Devices

Wearable-based evaluation of physical fatigue is carried out by means of multiple sensors. Generally, monitoring of heart rate (HR), which can be derived from either photoplethysmogram (PPG) or electrocardiogram, is included. This work presents an approach for classification of physical fatigue only on HR signals. Electromyogram (EMG) signals are instead employed for labeling, as the analysis of their power spectrum is considered a reference method for evaluating fatigue. In detail, the paper has the dual purpose of i) avoiding a multimodal analysis that would increase power consumption, and ii) defining a methodology easily applicable to the majority of commercial wearable devices, usually equipped with a PPG sensor. The experimental analysis is carried out for a set of HR and EMG signals acquired on subjects monitored by wireless devices. Specifically, physical fatigue felt by arms is detected during isometric muscle contractions. Then, the classification is implemented by comparing several machine learning algorithms. For all algorithms, the proposed approach for classification of physical fatigue reveals good performance in terms of Accuracy values, which are always higher than 85%. The boosting algorithm provides the highest value of Accuracy = 90.64% and the highest value of F1 = 89.78%. The obtained results, comparable to the literature values directly based on EMG feature extraction, prove the efficacy of the proposed approach, even more the performed classification is not simply intra-subject, but also inter-subject.