Identification of the optimal emotion recognition algorithm using physiological signals

In human-computer interaction researches, emotion recognition systems based on physiological signals have introduced. This study was to identify the optimal emotion recognition algorithm for classification of seven emotional states (happiness, sadness, anger, fear, disgust, surprise, and stress) using physiological signals. 12 college students participated in this experiment over 10 times. To induce each emotion, 10 emotional stimuli sets which had been tested their suitability and effectiveness, were used in experiment. Physiological signals, i.e. EDA, ECG, PPG, and SKT were acquired by MP100 Biopac system Inc. (USA) and AcqKnowledge (Ver. 3.8.1) software was used for analysis of the signals. Physiological signals were obtained prior to present emotional stimuli and while emotional stimuli were presented to participants. 28 features were extracted the acquired signals and analyzed for 30 seconds from the baseline and the emotional state. For emotion recognition, data subtracting baseline values from the emotional state applied to 5 machine learning algorithm, i.e. FLD, CART, SOM, Naïve Bayes and SVM. The result showed that an accuracy of emotion classification by SVM was 99.04% (which is the highest accuracy among used algorithms) and 30.14% by FLD (which is the lowest). This means that SVM is the optimal emotion recognition algorithm by data of this study. Our result can help emotion recognition studies lead to better chance to recognize not only basic emotion but also user´s various emotions, e.g., boredom, frustration, love, pain, etc., by using physiological signals. Also, it is able to be applied on many human-computer interaction devices for emotion detection.