Classification of Human Emotions from Physiological signals using Machine Learning Algorithms

Emotion recognition is one of the key steps towards emotional intelligence in advanced human-machine interaction. Recently, emotion recognition using physiological signals has been performed by various machine learning algorithms as physiological signals are important for emotion recognition abilities of human-computer systems. The purpose of this study is to classify three different emotional states (boredom, pain, and surprise) from physiological signals using several machine learning algorithms and to identify the optimal algorithms being able to classify these emotions. 217 subjects participated in this experiment. The emotional stimuli designed to induce three emotions (boredom, pain, and surprise) were presented to subjects and physiological signals were measured for 1 minute as baseline and for 1-1.5 minutes during emotional states. The obtained signals were analyzed for 30 seconds from the baseline and the emotional state and 27 parameters were extracted from these signals. For classification of three different emotions, machine learning algorithms of Decision tree, k-NN (k-nearest neighbor algorithm), LDA (linear discriminant analysis), and SVM (support vector machine) were done by using the difference values of signal parameters subtracting baseline from the emotional state. Classification accuracy using LDA was 74.9% and the result of emotion recognition using Decision Tree showed that accuracy to recognize all emotions was 67.8%. In analysis of k-NN and SVM, classification accuracy was 62.0%. The result of emotion recognition shows that LDA is the best algorithm being able to classify pain, surprise, and boredom emotions. This led to better chance to recognize other emotions except human basic emotions and to assist more accurate and greater understanding on emotional interactions between man and machine based on physiological signals.