J. Biomedical Science and Engineering, 2014, 7, 604-620
Published Online June 2014 in SciRes. http://www.scirp.org/journal/jbise
http://dx.doi.org/10.4236/jbise.2014.78061
How to cite this paper: Valenzi, S., Islam, T., Jurica, P. and Cichocki, A. (2014) Individual Classification of Emotions Using
EEG. J. Biomedical Science and Engineering, 7, 604-620. http://dx.doi.org/10.4236/jbise.2014.78061
Individual Classification of Emotions Using
EEG
Stefano Valenzi
1*
, Tanvir Islam
1*
, Peter Jurica
1
, Andrzej Cichocki
1,2
1
Advanced Brain Signals Processing Laboratory, RIKEN Brain Science Institute (RIKEN BSI), Wako, Japan
2
Systems Research Institute Polish Academy of Science, Warsaw, Poland
Email: stefano@brain.riken.jp
Received 20 April 2014; revised 30 May 2014; accepted 17 June 2014
Copyright © 2014 by authors and Scientific Research Publishing Inc.
This work is licensed under the Creative Commons Attribution International License (CC BY).
http://creativecommons.org/licenses/by/4.0/
Abstract
Many studies suggest that EEG signals provide enough information for the detection of human
emotions with feature based classification methods. However, very few studies have reported a
classification method that reliably works for individual participants (classification accuracy well
over 90%). Further, a necessary condition for real life applications is a method that allows, irres-
pective of the immense individual difference among participants, to have minimal variance over
the individual classification accuracy. We conducted offline computer aided emotion classification
experiments using strict experimental controls. We analyzed EEG data collected from nine partic-
ipants using validated film clips to induce four different emotional states (amused, disgusted, sad
and neutral). The classification rate was evaluated using both unsupervised and supervised learn-
ing algorithms (in total seven “state of the art” algorithms were tested). The largest classification
accuracy was computed by means of Support Vector Machine. Accuracy rate was on average 97.2%.
The experimental protocol effectiveness was further supported by very small variance among in-
dividual participants’ classification accuracy (within interval: 96.7%, 98.3%). Classification accu-
racy evaluated on reduced number of electrodes suggested, consistently with psychological con-
structionist approaches, that we were able to classify emotions considering cortical activity from
areas involved in emotion representation. The experimental protocol therefore appeared to be a
key factor to improve the classification outcome by means of data quality improvements.
Keywords
EEG, Human Emotions, Emotion Classification, Machine Learning, LDA
1. Introduction
Human emotion plays a critical role in perception, cognition and (social) behavior. Good management of emo-
*
These authors equally contributed to this work.