EEG Feature Extraction and Classification using Feed Forward Backpropogation Algorithm for Emotion Detection

Electroencephalography is a clinical technique which reads the scalp electrical activity from brain structures. The electroencephalogram (EEG) records the scalp surface using metal electrodes and conductive media[ 6], and inhibits lots of information to different emotional states and the recorded data is very complex. In the present work EEG signal is used to detect emotions at different situations. The real emotions are detected which acts as a real indicator described by the human subject. In this paper, the EEG signals are analyzed for feature extraction using "db4" wavelet using multilevel decomposition. 10 real human subjects real EEG samples are collected using standard International 10-20 Electrode placement system which is placed over an entire scalp of human subject and it is decomposed into 5 different EEG bands using Discrete wavelet Transform (DWT). Emotion detection is the task of emotional state like recognizing a person's happiness, fear, anger, confusion or deceit across both voice and non voice channels. Emotion detection is based on a set of conventional features which are extracted like energy, Power spectral density, from the EEG signals for classifying emotions and to identify the intensity level of different bands of EEG signal. Feature extraction results are obtained by using "Bior 5.5" and "db4" wavelet for signal decomposition and to obtain the accurate frequency bands and feature Classification is performed by obtaining an accuracy of 75% for normal patient and 65% for abnormal patient.