A Unified Dynamic Model for the Decomposition of Skin Conductance and the Inference of Sudomotor Nerve Activities

Abstract

In behavioral health informatics, inferring an individual's psychological state from physiological and behavioral data is fundamental. A key physiological signal in this endeavor is electrodermal activity (EDA), often quantified as skin conductance (SC), known for its sensitivity to a variety of psychological stimuli. Traditional methods to analyze skin conductance, such as the trough-to-peak method, often result in imprecise estimations due to overlapping skin conductance responses. While various mathematical models have been proposed to improve the analysis, many of them do not incorporate the tonic level in the dynamic system. This paper introduces a novel fourth order dynamic system to model the temporal dynamics of skin conductance, unifying both the tonic level and phasic response. Applied to a large dataset with over 200 participants, majority of the models achieved an $R^2$ value above 0.99. Furthermore, this work offers a unique three-component decomposition of skin conductance, shedding light on its temporal dynamics. Comparative evaluations highlight the model's capability to differentiate arousal levels and maintain an appropriate sparsity level for the estimated sudomotor nerve activities signal. The code of the proposed model and algorithm are available for open access.