Extracting the Features of Emotion from EEG Signals and Classify Using Affective Computing

The extraction of features, for the recognition of affective states through various means such as gestures of the body, facial images and electroencephalogram (EEG), is very important in affective computing. The brain-machine interface (BMI) using emotions, are used in medical robots, neuroergonomics, and auto-navigation and security systems. Emotions can be identified using analysis of scalp EEGs. The EEG data with audio-visual stimulus is collected and analyzed to extract the features of five emotions viz., happy, sad, fear, neutral and disgust. The raw EEG data is used to create the database, EEG_Amrita_emote. Features of EEG data are extracted using independent component analysis (ICA), and are classified using K Nearest Neighbor (KNN) algorithm. Cluster centroids are identified using k-Mean Clustering. The spectral energy of emotional activities in the brain is taken as one of the features. The EEG data is collected from male subjects of age group between 20 and 30. The locations of high intensity spectral energy is calculated for every emotion. The primary centroids of emotions are happy at (26.58, -99.97), neutral at (-69.18, 12.89), sad at (66.45, 29.52), fear at (74.22, -9.65) and disgust at (63.05, 38.68) respectively.