Discrimination in sentiments based on Galvanic skin response

Abstract

Feelings such as fear, anger, happiness, pain, love, and hatred, are arisen due to the surrounding environment or people, generally known as emotions. Emotions are the part of a human character that consists of their feelings, contrasting to their thoughts. Emotion detection is viral nowadays, as humans do their work from day to night without giving proper attention to their health. Because of that, stress increases day by day. The detection and prediction of human emotions are essential for controlling and handling critical situations. Interpersonal communication is about the language that is spoken and non-verbal indications like facial expression, hand and body gestures, tone of voice that are used for expressing their feeling and providing feedback. This research work proposes a system to classify human sentiments with the help of electrodermal activity and discriminate between different-different emotions viz. happy and relax, pain, etc. In this study, the GSR (Grove – GSR Sensor V1.2) sensor was used to observe 12 people. Data pre-processing done by using Moving Average Window and Normalization method. The analysis of data was computed in the time domain. The Supervised machine learning model used for classification viz. Support vector machine, Decision Tree, k-nearest neighbour. The study was conducted in CEERI's machine vision lab. The predicted accuracy for happy, relax, and pain activity was 91%, 97%, and 98%, respectively, with the help of a k-nearest neighbour model. The best suitable model to discriminate human sentiments is the KNN algorithm among all investigated supervised machine learning models.