Investigating the Generalizability of Physiological Traits of Nervousness
Authors: Emily Zhou, Mohammad Soleymani, Maja J. Matarić
Summary: Latest works have demonstrated the effectiveness of machine studying (ML) methods in detecting nervousness and stress utilizing physiological alerts, however it’s unclear whether or not ML fashions are studying physiological options particular to emphasize. To deal with this ambiguity, we evaluated the generalizability of physiological options which have been proven to be correlated with nervousness and stress to high-arousal feelings. Particularly, we study options extracted from electrocardiogram (ECG) and electrodermal (EDA) alerts from the next three datasets: Nervousness Phases Dataset (APD), Wearable Stress and Have an effect on Detection (WESAD), and the Repeatedly Annotated Indicators of Emotion (CASE) dataset. We purpose to know whether or not these options are particular to nervousness or common to different high-arousal feelings by means of a statistical regression evaluation, along with a within-corpus, cross-corpus, and leave-one-corpus-out cross-validation throughout situations of stress and arousal. We used the next classifiers: Assist Vector Machines, LightGBM, Random Forest, XGBoost, and an ensemble of the aforementioned fashions. We discovered that fashions skilled on an arousal dataset carry out comparatively effectively on a beforehand unseen stress dataset, and vice versa. Our experimental outcomes counsel that the evaluated fashions could also be figuring out emotional arousal as an alternative of stress. This work is the primary cross-corpus analysis throughout stress and arousal from ECG and EDA alerts, contributing new findings in regards to the generalizability of stress detection
