Emotional states classification from brain signals
Electroencephalography (EEG) signal analysis is very useful in the assessment of emotion mechanisms. With the study of emotions, it can gradually lead to integration with the Human-Computer Interaction (HCI) systems. In this project, these signals were generated by visual stimuli and are collected....
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
2013
|
Subjects: | |
Online Access: | http://hdl.handle.net/10356/55217 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Electroencephalography (EEG) signal analysis is very useful in the assessment of emotion mechanisms. With the study of emotions, it can gradually lead to integration with the Human-Computer Interaction (HCI) systems. In this project, these signals were generated by visual stimuli and are collected. The EEG signals were obtained using four electrode placements, according to 10-20 systems, from 5 subjects with 30 samples each. The time-domain and frequency-domain features of these signals are culled with a MATLAB algorithm.
There has been multiple studies pertaining to emotional recognition using EEG, and the application of Wavelet Transform in the fraternity is gaining more and more attention. We will study the results of the proposed Discrete Wavelet Transform (DWT) for feature extraction from EEG signals
The training of the EEG data has been done by past students working on vastly identical projects. We will observe the performances of the Kernel Extreme Learning Machine (ELM) compared to Support Vector Machine (SVM), and two other variants being – polynomial kernel with standard deviation, and polynomial kernel with maximum deviation. Using five machine learning databases, the performance of the proposed Discrete Wavelet Transform as feature extraction methods was analyzed.
Results from the experimental data have indicated that Kernel ELM is the best option as opposed to other SVM variants. Experimental results also showed that Kernel ELM constantly outperformed SVM in the emotion recognition from EEG signals. Discrete Wavelet Transform with Kernel ELM achieved 84.67% emotion recognition accuracy for two discrete emotions - “happy” and “sad”. |
---|