Quantitative Analysis for Emotion Recognition by Using EEG Signals

Electroencephalogram (EEG) signal is a recording of electrical activity across the scalps. It is a biological signal that often used as emotion recognition and has been widely adopted in medical, affective computing and others relevant field. The challenge in this research is to classify emotional states from the signals produced from the brain. A proper design of experiment required to make sure a correct emotion induced regarding the task given to the subject. Therefore, this research proposes a data acquisition protocol for inducing emotional states of the subject. Three type of stimuli (audio-visual, audio, visual) and three group of emotional states (Positive, Negative, Neutral) involve in the investigation. In the signal preprocessing stage, five features representing emotion signals are selected which is energy, mean, variance, entropy, and power. Moreover, for the emotion classification K-Nearest Neighbors (KNN) and Probabilistic Neural Network (PNN) are used as the classifier. From the result, energy and power features give the best performance among the five features selected. KNN produces the average accuracy 67.15%. Moreover, PNN produces the average accuracy of 64.87%.