Multilevel assessment of mental stress via network physiology paradigm using consumer wearable devices

Mental stress is a physiological condition that has a strong negative impact on the quality of life, affecting both the physical and the mental health. For such a reason, accurate measurements of stress level can be helpful to provide mechanisms for prevention and treatment. This paper proposes a procedure for the classification of different mental stress levels by using physiological signals provided by low invasive wearable devices. 17 healthy volunteers participated in this study. Three different mental states were elicited in them: a resting condition, a stressful cognitive state, and a sustained attention task. The acquired physiological signals were: a one lead electrocardiogram (ECG), a respiratory signal, a blood volume pulse (BVP), and 14 channels of a 10–20 electroencephalogram (EEG). For all subjects, 59 time series of 300 samples each were structured by including the RR series, the respiratory series, the pulse arrival time (PAT) series, and the delta, theta, alpha, beta power series of the 14 EEG channels. Different classifiers were implemented to assess the mental stress level starting from a pool of 3481 features computed from the aforementioned physiological quantities, using the Network Physiology paradigm. The highest achieved accuracy was 84.6%, from logistic regression and random forest classifiers, cross validated by mean of leave-one-person-out analysis. A further analysis was carried out to evaluate the classification accuracy using only cardio-respiratory signals, since the latter are more suitable to be used in real-life scenarios. In this case, the highest achieved accuracy was 76.5% obtained by the random forest classifier.