Real-Time EEG-Based Happiness Detection System

Table 1

EEG-based emotion recognition researches.

References	Year	Participant	Emotion	Stimulus	Feature	Temporal window	Classifier	Result	Real time
[10]	2006	4; subject-dependent	3 arousal classes	Picture	PSD	—	NB	58%	No
[11]	2008	26; subject-independent	4 classes (joy, anger, sadness, and pleasure)	Music	ASM	1 s	SVM	92.73%	No
[20]	2009	10; subject-dependent	2 valence classes	Picture	CSP	3 s	SVM	93.5%	No
[21]	2009	10; —	3 arousal classes	Recall	PSD	0.5 s	SVM	63%	No
[22]	2009	1; subject-dependent	3 classes (positively excited, negatively excited, and calm)	Picture	statistical features	—	QDA	66.66%	No
[23]	2009	3; subject-dependent	10 classes	Self-elicited	PSD	1 s	KNN	39.97–66.74%	No
[24]	2010	26; subject-independent	4 classes (joy, anger, sadness, and pleasure)	Music	ASM	1 s	SVM	82.29%	No
[25]	2010	6; subject-dependent	2 valence classes; 2 arousal classes	Music video	PSD	—	SVM	58.8% (valence); 55.7% (arousal)	No
[26]	2010	26; subject-dependent	4 classes (calm, happy, sad, and fear)	Picture and music	SOM	2 s	KNN	84.5%	No
[28]	2010	15; —	2 classes (calm-neutral and negatively excited)	Picture	HOS	2 s	SVM	82%	No
[29]	2010	12; subject-dependent	2 valence classes; 2 arousal classes	Sound	FD	—	threshold	—	Yes
[27]	2011	20; —	5 classes (happy, disgust, surprise, fear, and neutral)	video clip	Entropy	—	KNN	83.04%	No
[31]	2011	6; subject-dependent	2 valence classes	Movie clip	PSD	1 s	SVM	87.53%	No
[32]	2011	20; subject-independent	3 classes (boredom, engagement, and anxiety)	Game	PSD	—	LDA	56%	No
[33]	2011	5; subject-dependent	4 classes (joy, relax, sad, and fear)	Movie	PSD	1 s	SVM	66.51%	No
[34]	2011	11; —	3 valence classes	Picture	ASM	4 s	KNN	82%	No
[30]	2012	27; subject-independent	3 valence classes; 3 arousal classes	Video	PSD and ASM	—	SVM	57.0% (valence); 52.4% (arousal)	No
[35]	2012	32; —	2 valence classes; 2 arousal classes	Music video	PSD and ASM	—	NB	57.6% (valence); 62.0% (arousal)	No
[36]	2012	20; subject-dependent	5 classes (happy, angry, sad, relaxed, and neutral)	Picture	FD	—	SVM	70.5%	Yes
[37]	2012	5; subject-dependent	3 classes (positively excited, negatively excited, and calm)	Picture	HOC	—	KNN	90.77%	No
[38]	2012	4; —	2 valence classes; 2 arousal classes	Video clip	ASP	—	—	66.05% (valence); 82.46% (arousal)	No
[39]	2012	32; —	2 classes (stress and calm)	Music video	PSD	—	KNN	70.1%	No
[40]	2012	36; —	3 classes	Music video	PSD	—	ANN	—	Yes
[41]	2013	11; subject-independent	2 valence classes	Picture	PSD	4 s	SVM	85.41%	No

∗The feature, temporal window, and classifier shown in this table are the sets giving the best accuracy of each research.
Feature: Power Spectral Density (PSD), Spectral Power Asymmetry (ASM), Common Spatial Pattern (CSP), Higher Order Crossings (HOC), Self-Organizing Map (SOM), Higher Order Spectra (HOS), Fractal Dimension (FD), and Asymmetric Spatial Pattern (ASP).
Classifier: Support Vector Machine (SVM), Naïve Bayes (NB), Quadratic Discriminant Analysis (QDA), K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), Multilayer Perceptron (MLP), and Artificial Neural Network (ANN).