Brain signals - emotion states i

Abstract

Many past research works have been done to perform feature extraction and classification of human emotions based on brain signals. More commonly, many researches have chosen Discrete Wavelet Transform (DWT) coefficients as a feature vector as it gave a better accuracy compared to other features.

In this project, 6 subjects’ brain signals from four different electrode channels, each comprising of 30 samples were collected by conducting an experiment evoking emotions of both happy and sad for feature extraction and classification.

Brain signals have been preprocessed and analyzed to select the most consistent and prominent feature vector. Different feature vectors have been studied and classification methods have also been used to classify human emotions based on collected brain signals. Difference between FP1 and FP2 electrode channels were also studied for signals in different domains as part of a feature vector.

This project focuses on the use of Artificial Neural Network (ANN) classifier, together with features like DWT, Power Spectra Density (PSD) and time domain features comprising of mean and amplitudes for emotion classification. The best combination of feature vectors giving ANN classifier the highest classification accuracy of 81.8% comprises of discrete wavelet coefficients, power spectra coefficients and time domain amplitudes.