Differentiation of Cognitive Stress and Physical Activity Using Wearable ECG Derived Features

Real time stress detection during physical activity using wearable devices is challenging due to presence of extreme noises and baseline drift caused by movement. As, cognitive stress and physical activity causes changes in heart rhythm, it is important to accurately differentiate them for analysing their effects on cardiovascular health. This paper proposes a feature analysis using windowed heart rate variability (HRV) for distinguishing cognitive stress and physical activity. The ECG data obtained from wearable devices is further annotated with corresponding labels. The extracted data is preprocessed to remove baseline wandering and powerline interference by decomposing the signal using discrete wavelet transform for R-peak detection. The annotated preprocessed signals have been segmented into lOs window for R-peak detection to calculate the RR intervals. The time domain and frequency domain features are extracted and analyzed for distinguishing cognitive stress and physical activity. The frequency domain features show significance in distinguishing cognitive stress and physical activity during real time applications. The low-frequency to high-frequency power ratio shows a significant variation between cognitive stress and physical activity with a mean percentage deviation of 52%. Additionally, statistical ANOVA test shows that LF/HF power, normalized LF power and normalized HF power are statistically significant with p<0.05. Thus, frequency-based features can be effectively employed in wearable devices to analyze mental stress during daily activities.