MVAR and Causal Modeling of Relationship between Physiological Signals and Affective States

Abstract

This paper investigates the association and causality between multi-modal bio-signals measured using wearables, and the subjects’ affective state. The chosen methods are the multivariate auto-regressive model and Granger causality test. In particular, we focused on the state of stress and detected that respiratory, electrocardiogram, and accelerometer signals have the strongest associations with stress. These signals demonstrated the highest correlation values of 0.23, 0.2, and 0.18 respectively. The p-value for Granger causal F-test also shows a strong causal relationship between stress and the physiological signals.