ELECTRODERMAL ACTIVITY-BASED ANALYSIS OF EMOTION RECOGNITION USING TEMPORAL-MORPHOLOGICAL FEATURES AND MACHINE LEARNING ALGORITHMS

In this study, we evaluated the performance of tonic and phasic components of Electrodermal activity (EDA) using machine learning algorithms for accurately recognizing emotions. The EDA signals considered for this study were obtained from Continuously Annotated Signals of Emotion (CASE) dataset. Initially, we pre-processed and decomposed the EDA into tonic and phasic components using cvxEDA method. Further, we extracted the temporal and morphological features from both tonic and phasic. Finally, we tested the performance of various combinations of features using machine learning algorithms such as logistic regression, support vector machine (SVM), and random forest. Our results revealed that the tonic contributes significant information for emotional state classification. Further, the temporal features of the phasic were able to discriminate most of the emotions [Formula: see text]. In particular, the scary emotion was well discriminated against other emotions. Results of classification revealed that SVM performed best in classifying emotional states. The results of our process pipeline, which incorporated tonic, temporal features, and SVM, showed impressive classification performance with average accuracy, sensitivity, specificity, precision, and f1-score of 78.96%, 57.92%, 85.97%, 62.32%, and 56.48%, respectively. Our findings indicate that our proposed models could potentially be used to detect the positive and negative emotions in healthcare settings.