Emotions Detection based on a Single-electrode EEG Device

Royl

an Quesada-Tabares, Alberto J. Molina-Cantero, Isabel M. G

omez-Gonz

alez,

Manuel Merino-Monge, Juan A. Castro-Garc

ıa and Rafael Cabrera-Cabrera

ETS Ingenier

ıa Inform

atica, Departamento de Tecnolog

ıa Electr

onica, Universidad de Sevilla, Spain

Keywords:

Emotions, Signal Processing, Single EEG Electrode, Classiﬁcation Analysis.

Abstract:

The study of emotions using multiple channels of EEG represents a widespread practice in the ﬁeld of research

related to brain computer interfaces (Brain Computer Interfaces). To date, few studies have been reported in

the literature with a reduced number of channels, which when used in the detection of emotions present results

that are less accurate than the rest. To detect emotions using an EEG channel and the data obtained is useful

for classifying emotions with an accuracy comparable to studies in which there is a high number of channels,

is of particular interest in this research framework. This article uses the Neurosky Maindwave device; which

has a single electrode to acquire the EEG signal, Matlab software and IBM SPSS Modeler; which process

and classify the signals respectively. The accuracy obtained in the detection of emotions in relation to the

economic resources of the hardware dedicated to the acquisition of EEG signal is remarkable.

1 INTRODUCTION

There are two main theories about the nature of emo-

tions. One of them posits the existence of a relatively

low number of basic emotions (families of emotions)

that are universal for all human beings (Ekman et al.,

1987; Levenson, 2011). At least six families have

been proposed: happiness, sadness, repulsion, anger,

fear and surprise. The dimensional theory, on the

other hand, considers that emotions are represented in

an N-dimensional space, where two of the coordinate

axes would explain most of the emotional variations.

These axes are called Valence and Arousal (Russell

and Barrett, 1999). Valence is related to pleasure and

varies from negative values (very unpleasant) to pos-

itive values (very pleasant). Arousal is related to the

intensity of emotion, ranging from very low to very

high. The two theories of emotions are not contra-

dictory to each other. In fact the six basic emotions

can be characterized according to their valence and

arousal.

The theory of basic emotions includes, in turn, a

locationist model that assumes that each emotional

category starts from a speciﬁc place of the brain and

body. Speciﬁcally, fear is located in the amygdala; the

feeling of repulsion, in the insula; anger in the orbito-

frontal cortex (OFC) and sadness in the anterior

cortex of the cingulate (ACC) (Vytal and Hamann,

2010). Dimensional theory is included within the so-

called constructionist model where it is asserted that

emotions are psychological events that emerge from

basic physiological operations, which are not speciﬁc

to emotions. In (Lindquist et al., 2012) the authors

propose a model with four components: core affect,

a body sensory input that is experienced as pleasant /

unpleasant (valence) with some degree of excitation;

conceptualization, which links the body sensations

with previous experiences to endow them with mean-

ing; emotional words, used as support of emotional

categories that are not clearly differentiable from the

sensitive point of view; and executive attention, which

focuses on some of the incoming stimuli. Some neu-

roimaging results have corroborated that, unlike what

the locationist model predicts, any region that was ac-

tivated during a basic emotion, was also activated for

at least one other emotion (Lindquist et al., 2012).

This suggests the existence of neural networks that

interact with each other to generate the emotions, in-

stead of precise places (locationist model). For exam-

ple the amygdala is recruited for both fear and repul-

sion, so it takes different functionalities depending on

the neural network that uses it.

There is no one theory that dominates, taking into

account the results of neuroimaging, which can be in-

terpreted differently according to the procedure used

for the treatment of the data. (Hamann, 2012) summa-

rizes the existing controversy pointing out that in the

future, the analysis of animal models and studies on

patients with brain injuries should be undertaken,

Quesada-Tabares, R., Molina-Cantero, A., Gómez-González, I., Merino-Monge, M., Castro-García, J. and Cabrera-Cabrera, R.

Emotions Detection based on a Single-electrode EEG Device.

DOI: 10.5220/0006476300890095

In Proceedings of the 4th Inter national Conference on Physiological Computing Systems (PhyCS 2017), pages 89-95

ISBN: 978-989-758-268-4

which have been reported as tending more towards a

locationist theory.

Emotions can be determined in several ways:

through the analysis of gestures (facial or other), texts,

speech (Liu et al., 2011), as well as the activity of var-

ious physiological variables. The electroencephalo-

gram (EEG) is one of these variables. In this arti-

cle we analyze the EEG and make use of the two-

dimensional model.

There are a large number of devices with which

EEG activity can be measured. These vary in price

and how the measurement obtained is transmitted

(wired or wireless). The use of a wireless device was

opted for due to the advantages in mobility and adapt-

ability it offers. Some devices that are on the market

with these characteristics are: Emotiv (emo, ), Neu-

rosky Mindwave (emo, ) and Enobio(eno, ). Emotiv

offers better results than the Neurosky Mindwave if it

is used for the evaluation of cognitive processes (Das

et al., 2014). Instead of the support offered, usabil-

ity and its price so competitive is selected to make

this study the Neurosky Mindwave. This has been

widely used by the scientiﬁc community for the de-

velopment of various applications such as the detec-

tion of sleepiness (Van Hal et al., 2014), level of at-

tention (Liu et al., 2013), stress (Crowley et al., 2010;

Maki et al., 2012), and so on. It is a device that of-

fers developers and researchers the possibility to treat

the measured signal, but it also comes integrated with

a system that processes and delivers characteristics of

the post-processing to the user, which will not be used

in our case.

For the study and comprehension of the EEG sig-

nal, the analysis of the bands is widely used. An-

other feature used to study the EEG signal is the frac-

tal dimension (Wang and Sourina, 2013; Siamaknejad

et al., 2014) (in this study the Higuchi algorithm was

used (Cervantes-De la Torre et al., 2013)). A fractal

dimension close to 2 indicates that the signal is very

complex, however a value close to 1 means that the

signal is close to a line.

This study aims to analyze the characteristics of

EEG signal, study the statistical behavior and make

a classiﬁcation of the emotional states using a set of

images.

2 METHODOLOGY

2.1 Materials

To acquire the EEG signal we used Neurosky Mind-

wave. This device has an electrode placed at Fp1, ac-

cording to the standard 10-20 system, and a ﬁxed

Figure 1: Valence and arousal values of the selected pic-

tures.

sampling frequency of 512Hz with a bluetooth inter-

face. The data is read, saved and processed using

Matlab 8.4.0.150421 (R2014b). To study the signal

features we used the IBM SPSS Modeler, this soft-

ware is a set of tools of data mining that allows for

the quick development of predictive models, which

offers a wide variety of modeling methods from au-

tomated learning, artiﬁcial intelligence and statistics

(Corporation, 2012).

2.2 Experimentation

Seven people took part in the experiment. Their av-

erage age was 29.85 with a standard deviation of

8.97. The experiment consists of displaying 60 pic-

tures of the IAPS (International Affective Picture Sys-

tem), formed by three different groups of valence and

arousal pairs (Figure 1), The ﬁrst group has 2.306

± 0.43 valence and 6.1890 ± 0.04 arousal, the sec-

ond 5.063 ±0.24 valence and 3,020 ± 0.02 arousal,

while the third 6,921 ± 0.032 valence and 4,551 0.02

arousal.

Figure 2 depicts the timeline of the experiment.

Before starting it, the SAM test was applied (Self-

assessment manikin). This test allows a quick self-

assessment of each participant indicating their initial

values of valence and arousal. Then, the sixty IAPS

pictures were shown randomly, each displayed for 6s,

following the same procedure as in (Aftanas et al.,

2001). Between each picture a resting picture with

a black background and a gray cross in the center was

shown for 4s (Hosseini and Naghibi-Sistani, 2011).

This time lag between IAPS pictures reduces the over-

lapping effect on the EEG signal.

A Matlab software was developed to show the pic-

tures, record the EEG data sent by Neurosky and in-

troduce time marks to build EEG epochs associated

with each IAPS picture and resting period. All the

sessions were recorded by a WebCam to contrast any

possible anomaly in the signal. Finally, participants

completed a new SAM test for each picture.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

2.3 Signal Processing

The EEG signal was segmented using windows of 512

samples (1-s) with a hop size of 64 samples (overlap-

ping 87.5 %). A procedure has been developed for the

automatic analysis of each epoch, in order to identify

whether it contains a valid EEG signal, or whether it is

contaminated by any type of artifact. These artifacts

may have different sources: blinks, winks, eye move-

ments, motion artifact, or muscle activity (EMG). In

our case, the main source of artifacts have an ocular

origin, because of the position of the electrode, al-

though what is also important is the electrical activity

of the frontal and temporal muscles and the artifacts

due to the movement of the electrode.

The pre-processing of the signal looks to identify

possible artifacts in each epoch. To accomplish this,

we have used two features: the difference between

the maximum and minimum sample value (MinMax),

and the total energy (ESF) of the signal after applying

a Savitzky-Golay lowpass ﬁlter (order 2 and length

35) (Schafer, 2011). Figure 3 shows the feature space.

Epochs containing muscular activity have values of

the MinMax feature that are similar or a bit higher

than those of the epochs with only EEG, but with

more energy from the ﬁltered signal (ESF). Blinking

or EEG-only windows have similar values in the ESF

feature but differ in MinMax. Finally, windows with

motion artifact contain values of these features that

surround those obtained by other types of artifacts.

For all these reasons, the use of thresholds (maxi-

mum and minimum) of each dimension of the feature

space has been proposed, to limit and facilitate the

automatic detection of valid EEG container segments

and blinks (as shown in Figure 3) with an accuracy of

96 % and 98 % respectively. The method followed is

conservative in the selection of valid epochs, reducing

the number of false positives at the cost of increasing

false negatives.

While motion and EMG artifact are infrequent, the

ocular ones are not. There are techniques for remov-

ing ocular artifacts from the EEG signal. It is known

that ocular artifacts inﬂuence, fundamentally, the low-

est energy bands (δ, θ and part of α) so their non-

elimination of analysis windows could distort the re-

sults. One of the most used techniques for elimination

of these artifacts is based on the analysis of indepen-

dent components (Stone, 2004), but, for their applica-

Figure 2: Experimental sequence.

Figure 3: Feature spaces values for motion artifacts (pink),

muscular activity (blue), blinks (green) and EEG (black).

The selected areas to identify EEG and EEG+blinks epochs

are also shown.

tion, at least two EEG channels are required, which do

not exist in our case. However, in (Szibbo et al., 2012)

a technique has been developed to eliminate such arti-

facts in a single channel. It is based on applying a low

pass ﬁlter of the Savitzky-Golay type, with the same

characteristics as the one used for the calculation of

the ESF feature. Therefore, if an epoch is identiﬁed

as a blinking container, this type of ﬁlter is applied

before proceeding to the study in frequency.

The epochs containing no artifacts or only blinks

which have been previously removed are then win-

dowed with a Hamming function to reduce spectral

leakage. Then the squared fast Fourier transform

(FFT) is applied to each segment to obtained the typi-

cal energy bands:δ, θ, α and β. The fractal dimension

is also calculated, so a total of 5 values per epoch were

obtained.

2.4 Analysis

The averages of energy bands and fractal dimension

of the epochs contained in every 6s of picture were

obtained. From those averages the relation θ/β and

θ/α was also included in the posterior analysis. This

gives a total of 3x 20 x 7 numerical values for each

set of images, picture and subject. Then we apply two

different types of analysis: statistical and classiﬁca-

tion. The ﬁrst looked for signiﬁcant differences for

each feature among the 3 sets of pictures. The second

analysis sought to identify the accuracy in the detec-

tion of each set using a classiﬁer based on a decision

tree.

Emotions Detection based on a Single-electrode EEG Device

Table 1: Averaged values of each feature, subject and set of pictures. Values are ×10

Subject Set1 Set2 Set3

δ θ α β

hfd δ θ α β

hfd

1 6,04 5,12 2,3 3,21 1,59 2,22 1,58 3,36 3,65 1,95 3,08 1,18 1,86 1,59 4,02 3,98 1,96 2,89 1,37 2,02 1,58

2 2,06 2,49 2,63 3,9 0,63 0,94 1,66 2,1 2,59 2,95 4,08 0,63 0,87 1,65 2,19 3,26 2,85 3,77 0,86 1,14 1,65

3 4,32 5,48 4,37 8,64 0,63 1,25 1,57* 4,58 5,83 4,98 8,61 0,67 1,17 1,55* 4,06 5,63 4,67 8,3 0,67 1,2 1,58*

4 2,96 3,29 1,77 5,34 0,61 1,85 1,64 2,15 1,91 1,1 5,11 0,37 1,73 1,64 2,45 2,2 1,29 5,14 0,42 1,69 1,65

5 2,18 2,64 1,56* 3,81 0,69 1,69 1,68 1,61 2,09 1,1* 3,51 0,59 1,88 1,68 2,29 2,88 1,47* 3,6 0,8 1,95 1,67

6 4,68 4,94 3,09 4,5 1,09 1,6 1,58* 4,54 4,64 3,09 4,73 0,98 1,5 1,6* 3,56 4,4 3,05 4,76 0,92 1,44 1,63*

7 1,11 1,24 0,69* 4,51 0,27 1,78 1,71* 2,62 1,9 1,32* 4,53 0,41 1,43 1,68* 2,13 2,53 1,64* 5,07 0,49 1,53 1,69*

2.4.1 Study 1

The mathematical tool used to perform the ﬁrst anal-

ysis was ANOVA (Analysis of Variance) applied to

each person and to the seven features ordered by set

of images, in order to check if, at least one of the sets

of images, is signiﬁcantly different from the rest.

2.4.2 Study 2

The second analysis was performed using the classi-

ﬁcation algorithm C5.0, which generates a decision

tree. The division of the samples was based on the

node that offers the maximum gain of information in

each level and allows several divisions in more than

two subgroups iteratively until it arrives to perform

Divisions that do not have a signiﬁcant impact on the

model, which are discarded. The 7 features are taken

as input data for each image and have three nominal

values (the three sets in which the images have been

grouped). Then for each person the accuracy of the

correspondence established between each image and

its respective group is calculated. This is done using

IBM SPSS Modeler software, with the option of par-

titioning the data, to ensure that they have not used

the same information as in the model generation.

3 RESULTS

Table 1 shows the absolute values of the features ob-

tained for Study 1, grouped according to a sub-set

of images and a subject. The values that have been

shown to be statistically signiﬁcant, with a value of

p < 0.5, are emboldened. Subjects 3, 6 and 7 ob-

tained values signiﬁcant for the fractal dimension and

subjects 5 and 7 in the α band. However, there is not

a clear relationship between features’ behavior and

changes on valence and arousal. For example, for

subject 3, the fractal dimension has a behavior con-

cave with the sub-set of images (from set1 to set3).

However, for the subject 6, the same feature, has a

growing monotone behavior. Subject 7 has a concave

behavior too although between Set2 and the Set3 there

are no changes in the fractal dimension.

Subjects 5 and 7 show signiﬁcant changes in the

α band. However, like with the fractal dimension,

there is not a regular pattern of behavior of this feature

when we vary the sub-set of images. For example,

while subject 7 shows a behavior growing monotone

in the α band (from Set1 towards Set3), subject 5, has

a concave behavior.

In Table 2 the results of classiﬁcation by subject

can be seen, which are around 81% accurate.

Table 2: Accuracy in detecting emotional states.

Subject Accuracy

1 78,33%

2 78,33%

3 76,67%

4 73,33%

5 91,67%

6 81.67%

7 85%

Mean 80.71%

4 DISCUSSION

To facilitate the understanding of the results, the

Arousal and Valence results according to IAPS (Fig-

ure 1) and the most signiﬁcant features (hfd and

α), depending on the groups of selected images, are

shown (See ﬁgures 4 and 5).

4.1 Signiﬁcant Features Analysis

The fractal dimension is related to the complexity of

the EEG signal, which is increased in neural activa-

tion processes. In (Colibazzi et al., 2010) an experi-

ment was carried out with neuroimaging to ﬁnd out

which areas of the brain are activated by know vi-

sual stimuli categorized with different values of va-

lence and arousal. It is suggested that a stimulus with

positive valence decreases activity in the right hemi-

sphere, mainly in r-dLPFC and AMCC areas. In con-

trast, for a stimulus with positive excitement, activity

in the left hemisphere present a growing monotony,

fundamentally in thalamus and amygdala. Since the

measurement system mainly picks up the activity that

is located in the left hemisphere, the signal must be

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Set

1 1.5 2 2.5 3

Arousal

Set

1 1.5 2 2.5 3

Valence

Figure 4: Valence and Arousal variation of the three set of

selected pictures.

1 1.5 2 2.5 3

1.55

1.6

1.65

1.7

1.75

hfd

Subject 3

Subject 6

Subject 7

1 1.5 2 2.5 3

1.5

1.55

1.6

1.65

Alpha Band

Subject 5

Subject 7

Figure 5: Variation of the two statistically signiﬁcant fea-

tures for each set and subject.

more inﬂuenced by the arousal than by the valence.

Moving from set1 to set2 it is shown a decrease in

arousal and an increase in valence is shown, but the

fractal dimension for subjects 3 and 7 decreases as a

consequence of that reduction of activity in the left

hemisphere. When passing from set2 to set3, both

valence and arousal increase, growing the fractal di-

mension, as occurs with subjects 3, 6 and 7.

Another neuroimaging study (Nielen et al., ) de-

termines that the zones that are activated during

changes of valence and arousal differ from the pre-

vious one, but which also justify, to some extent, the

changes obtained in the fractal dimension. In it, the

arousal increase is correlated with the activity’s in-

crease in the areas associated with the medium tem-

poral gyrus (mT) and vLPFC. The behavior of valence

is more complex, as for negative values of this there is

a reverse neuronal activity with valence in the LPFC

and direct with it for positive values in the orbito-

frontal and mT regions. All activation areas can inﬂu-

ence the signal sensed with greater or lesser weight.

In any case, the change from set1 to set2 causes a gen-

eralized decrease of the prefrontal and temporal act-

ivity that could justify the decrease of the fractal di-

mension. The step from set2 to set3 shows increase

of activity in orbito-frontal (valence) and mT (va-

lence and arousal). There may be some compensation

in the measured signal as a function of the valence,

as both the sensor Fp1, affected by the orbitofrontal

zone, such as reference (located near the temporal

lobe) could counteract its effect, leaving only the de-

pendence of the arousal on the reference sensor, and,

therefore, in the recorded signal. This could justify

the increase of the fractal dimension.

The behavior of the other feature, the α band, does

not correspond to any valid fact in the revised scien-

tiﬁc literature. For example, an EEG system with 19

electrodes was used in (Valenza et al., 2015) to an-

alyze the inﬂuence of valence and arousal on power

bands. The authors used pictures of IAPS classiﬁed

according to two classes of valence (positive and neg-

ative) and four levels of arousal. It was concluded

that there were signiﬁcant changes in the bands θ and

β in the PFC and the parietal zone for intermediate

values of arousal. In (Aftanas et al., 2001), using 62

electrodes, they found signiﬁcant differences in the

band θ in the anterior temporal zone. There are ef-

fects of lateralization, since for negative valences, it

was observed a greater synchronization (activation) in

the left hemisphere, whereas for positive valences, the

same effect was observed in the opposite hemisphere.

4.2 Classiﬁcation Results

In (Yoon et al., 2013) a method was proposed to iden-

tify four emotional states: active, commitment, plea-

sure and neutral using the same system as in this pa-

per, but instead of being based on the pure signal,

it is based on two parameters: levels of mediation

and attention. The achieved classiﬁcation accuracy

was 66% with data from 42 participants. In (Brown

et al., 2011) the percentage of classiﬁcation was 82%,

higher than that obtained in this last work, but based

on the use of 9 electrodes, located mainly in the pre-

frontal area and in both hemispheres and based on

the ratio of the α band between the sensors located

in symmetrical positions between both hemispheres.

In (Chanel et al., 2007) 9 bands distributed between

[4, 20] Hz were used as features for training two

classiﬁers (LDA -Linear Discriminant Analysis- and

SVM -Support Vector Machine-) and three classes

(excitated, little excited, neutral) for arousal and two

classes for valence (positive, negative). The best re-

sults were obtained with SVM with 67% for arousal

and 76% for valence. In (Liu et al., 2011; Sourina and

Liu, 2011) , the fractal dimension was used to classify

states of excitation and valence with three electrodes.

Emotions Detection based on a Single-electrode EEG Device

One of them, located at FC6, was used for the arousal

and the difference between the AF3-F4 electrodes, for

the valence. Using three states for each variable (pos-

itive, neutral and negative) they achieved an accuracy

ranging from 70% up to 100% in some cases. In (Bos

et al., 2006) power bands were used as features for the

classiﬁer and a reduced number of electrodes. In par-

ticular, for arousal it was observed that the best clas-

siﬁcation result, 97.4 was with β band between the

electrodes F3 / F4, while that for valence, the result

was of the 92.3% in the same F3/F4 or in Fpz.

4.3 Affective Assessment

Finally, ﬁgure 6 shows the averages of reported

arousal and valence for each set of pictures per subject

(circle) and the averages for all subjects (*).

Figure 6: Mean experimental valence and arousal values for

each set of pictures and subject.

Although it can be seen that averages per sub-

ject approximate to the values of the IAPS (except

for set1, whose valence is greater and the arousal is

lower), the dispersion of the values reported for each

of the 20 pictures is quite higher than those indicated

in the IAPS database. These variations could be due

to cultural factors. In a study carried out in Spain with

more than 800 people, an adaptation of the values of

valence and excitation of the IAPS have been pro-

posed (Molt

o et al., 1999). The results differ slightly

from values reported in the IAPS, but, in the case of

this work, the dispersion obtained for each set was

even larger than the collection made by the Spanish

adaptation. This could justify, in large part, why sig-

niﬁcant variations among the measured features for

the different sets of pictures have not been obtained.

5 CONCLUSIONS

The study shows the way in which an EEG channel

is used and can perceive the response of neuronal ac-

tivity to stimuli; in our case, visual stimuli.The va-

lidity of using a low-cost commercial device such as

Neurosky MindWave for the acquisition of the signal

is checked. After implementing a classiﬁcation algo-

rithm, emotions are detected with an average accuracy

of 81% of total stimuli.This value exceeds the results

obtained in most of the studies reported in the liter-

ature, either for those who use a reduced number of

channels of EEG or for those who make the measure-

ment with a considerable number of electrodes.

There seems to be a direct correlation between

the signal complexity and the arousal; on the con-

trary, having a single electrode does not have sufﬁ-

cient information to give any conclusions about the

valence.There are variations between the IAPS data

and those reported by the people in the SAM test ap-

plied at the end of the experiment. It could inﬂuence

in the correspondence between the features expected,

taking as reference the IAPS and those experienced

by the subjects.

ACKNOWLEDGMENTS

We would like to thank all of those involved in the

realization of this study and the anonymous review-

ers who helped us improve this document with their

comments.

REFERENCES

Aftanas, L., Varlamov, A., Pavlov, S., Makhnev, V., and

Reva, N. (2001). Affective picture processing: event-

related synchronization within individually deﬁned

human theta band is modulated by valence dimension.

Neuroscience letters, 303(2):115–118.

Bos, D. O. et al. (2006). Eeg-based emotion recognition.

The Inﬂuence of Visual and Auditory Stimuli, pages

1–17.

Brown, L., Grundlehner, B., and Penders, J. (2011). To-

wards wireless emotional valence detection from eeg.

In 2011 Annual International Conference of the IEEE

Engineering in Medicine and Biology Society, pages

2188–2191. IEEE.

Cervantes-De la Torre, F., Gonz

alez-Trejo, J., Real-

Ram

ırez, C., and Hoyos-Reyes, L. (2013). Fractal

dimension algorithms and their application to time

series associated with natural phenomena. In Jour-

nal of Physics: Conference Series, volume 475, page

012002. IOP Publishing.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Chanel, G., Ansari-Asl, K., and Pun, T. (2007). Valence-

arousal evaluation using physiological signals in an

emotion recall paradigm. In 2007 IEEE International

Conference on Systems, Man and Cybernetics, pages

2662–2667. IEEE.

Colibazzi, T., Posner, J., Wang, Z., Gorman, D., Gerber,

A., Yu, S., Zhu, H., Kangarlu, A., Duan, Y., Rus-

sell, J. A., et al. (2010). Neural systems subserving

valence and arousal during the experience of induced

emotions. Emotion, 10(3):377.

Corporation, I. (2012). Manual de usuario de IBM SPSS

Modeler 15. page 280.

Crowley, K., Sliney, A., Pitt, I., and Murphy, D. (2010).

Evaluating a brain-computer interface to categorise

human emotional response. In 2010 10th IEEE In-

ternational Conference on Advanced Learning Tech-

nologies, pages 276–278. Ieee.

Das, R., Chatterjee, D., Das, D., Sinharay, A., and Sinha,

A. (2014). Cognitive load measurement-a methodol-

ogy to compare low cost commercial eeg devices. In

Advances in Computing, Communications and Infor-

matics (ICACCI, 2014 International Conference on,

pages 1188–1194. IEEE.

Ekman, P., Friesen, W. V., O’Sullivan, M., Chan, A.,

Diacoyanni-Tarlatzis, I., Heider, K., Krause, R.,

LeCompte, W. A., Pitcairn, T., Ricci-Bitti, P. E., et al.

(1987). Universals and cultural differences in the

judgments of facial expressions of emotion. Journal

of personality and social psychology, 53(4):712.

Hamann, S. (2012). Mapping discrete and dimensional

emotions onto the brain: controversies and consensus.

Trends in cognitive sciences, 16(9):458–466.

Hosseini, S. A. and Naghibi-Sistani, M. B. (2011). Emo-

tion recognition method using entropy analysis of eeg

signals. International Journal of Image, Graphics and

Signal Processing, 3(5):30.

Levenson, R. W. (2011). Basic emotion questions. Emotion

Review, 3(4):379–386.

Lindquist, K. A., Wager, T. D., Kober, H., Bliss-Moreau, E.,

and Barrett, L. F. (2012). The brain basis of emotion: a

meta-analytic review. Behavioral and Brain Sciences,

35(03):121–143.

Liu, N.-H., Chiang, C.-Y., and Chu, H.-C. (2013). Recog-

nizing the degree of human attention using eeg signals

from mobile sensors. Sensors, 13(8):10273–10286.

Liu, Y., Sourina, O., and Nguyen, M. K. (2011). Real-

time eeg-based emotion recognition and its applica-

tions. In Transactions on computational science XII,

pages 256–277. Springer.

Maki, Y., Sano, G., Kobashi, Y., Nakamura, T., Kanoh,

M., and Yamada, K. (2012). Estimating subjective

assessments using a simple biosignal sensor. In Soft-

ware Engineering, Artiﬁcial Intelligence, Networking

and Parallel & Distributed Computing (SNPD), 2012

13th ACIS International Conference on, pages 325–

330. IEEE.

Molt

o, J., Monta

es, S., Gil, R. P., Segarra, P., Verchili,

M. C. P., Ir

un, M. P. T., Ram

ırez, I., Hern

andez, M.,

anchez, M., Fern

andez, M., et al. (1999). Un m

etodo

para el estudio experimental de las emociones: el in-

ternational affective picture system (iaps). adaptaci

espa

nola. Revista de psicolog

ıa general y aplicada:

Revista de la Federaci

on Espa

nola de Asociaciones

de Psicolog

ıa, 52(1):55–87.

Nielen, M., Heslenfeld, D., Heinen, K., Van Strien, J., Wit-

ter, M., Jonker, C., and Veltman, D. Distinct brain sys-

tems underlie the processing of valence and arousal of

affective pictures. Brain and Cognition, 71(3).

Russell, J. A. and Barrett, L. F. (1999). Core affect, pro-

totypical emotional episodes, and other things called

emotion: dissecting the elephant. Journal of person-

ality and social psychology, 76(5):805.

Schafer, R. (2011). What is a savitzky-golay ﬁlter? [lec-

ture notes]. Signal Processing Magazine, IEEE,

28(4):111–117.

Siamaknejad, H., Loo, C. K., and Liew, W. S. (2014).

Fractal dimension methods to determine optimum eeg

electrode placement for concentration estimation. In

Soft Computing and Intelligent Systems (SCIS), 2014

Joint 7th International Conference on and Advanced

Intelligent Systems (ISIS), 15th International Sympo-

sium on, pages 952–955.

Sourina, O. and Liu, Y. (2011). A fractal-based algorithm of

emotion recognition from eeg using arousal-valence

model. In BIOSIGNALS, pages 209–214.

Stone, J. V. (2004). Independent Component Analysis. A

tutorial introduction. MIT Press.

Szibbo, D., Luo, A., and Sullivan, T. J. (2012). Removal of

blink artifacts in single channel eeg. In 2012 Annual

International Conference of the IEEE Engineering in

Medicine and Biology Society, pages 3511–3514.

Valenza, G., Greco, A., Lanata, A., Gentili, C., Menicucci,

D., Sebastiani, L., Gemignani, A., and Scilingo, E. P.

(2015). Brain dynamics during emotion elicitation in

healthy subjects: An eeg study. In 2015 AEIT Interna-

tional Annual Conference (AEIT), pages 1–3. IEEE.

Van Hal, B., Rhodes, S., Dunne, B., and Bossemeyer, R.

(2014). Low-cost eeg-based sleep detection. In En-

gineering in Medicine and Biology Society (EMBC),

2014 36th Annual International Conference of the

IEEE, pages 4571–4574. IEEE.

Vytal, K. and Hamann, S. (2010). Neuroimaging support

for discrete neural correlates of basic emotions: a

voxel-based meta-analysis. Journal of Cognitive Neu-

roscience, 22(12):2864–2885.

Wang, Q. and Sourina, O. (2013). Real-time mental arith-

metic task recognition from eeg signals. Neural Sys-

tems and Rehabilitation Engineering, IEEE Transac-

tions on, 21(2):225–232.

Yoon, H., Park, S.-W., Lee, Y.-K., and Jang, J.-H. (2013).

Emotion recognition of serious game players using

a simple brain computer interface. In ICT Con-

vergence (ICTC), 2013 International Conference on,

pages 783–786. IEEE.

Emotions Detection based on a Single-electrode EEG Device