CHARACTERIZATION OF DICHOTOMOUS EMOTIONAL STATES USING ELECTRODERMAL ACTIVITY BASED GEOMETRIC FEATURES

In this work, an attempt has been made to classify dichotomous emotional states using Electrodermal activity (EDA) and geometric features. For this, the annotated happy and sad EDA is obtained from the online public database. The EDA is subjected to discrete Fourier transform, and Fourier coefficients in the complex plane are obtained. The envelope of the complex plane is identified using the α-shape method. Five geometric features, namely center of gravity, eccentricity, convexity, rectangularity, and convex hull area are computed from the envelope and statistical analysis is performed. Two machinelearning algorithms, namely random forest (RF) and support vector machine, are considered for the classification. The results show that the proposed approach is able to classify the dichotomous emotional states. The rectangularity feature is found to be distinct and shows a statistically significant difference between the happy and sad emotional states (p<0.05). The RF classifier yields the highest F-m and AUC of 87.8% and 93.8%, respectively in differentiating emotional states. Thus, it appears that the proposed method could be used to understand the neurological, psychiatric, and biobehavioral mechanisms associated with happy and sad emotional states.