Investigating the Effects of Two-Class Categorical Emotion Classification Through Electrodermal Activity and Machine Learning

Abstract

Wearable emotion recognition using physiological signals gains attention due to its non-intrusive nature, yet effectively using physiological data remains a challenge. In our study, we crafted machine learning models for two-class categorical emotion recognition systems employing Electrodermal Activity (EDA). Initially, EDA signals were obtained from the continuously annotated signals of emotion dataset then pre-processed and decomposed the EDA into phasic signals using the convex optimization approach to EDA. We further extracted temporal features from the phasic signal. Finally, all extracted features and selected features through recursive feature elimination with cross-validation (RFECV) were applied to machine learning algorithms, such as logistic regression, support vector machine, random forest (RF), extreme gradient boosting, and multi-layer perceptron, to classify various two-class categorical emotions. The results of our process pipeline, encompassing phasic signals, RFECV-selected features, and RF, showcased impressive classification performance between the scary and boring emotion classes with an accuracy of 84.17%. Our findings suggest that the models we've proposed hold potential for application in various healthcare settings, effectively detecting both positive and negative emotions and contributing to the enhancement of individual well-being. © 2023 IEEE.