An Affective-computing Approach to Provide Enhanced Learning

Analytics

Javier Dorado Chaparro

, Rub

en Cantarero Navarro

, Ana Rubio Ruiz

Jes

us Fern

andez-Bermejo Ruiz

, Xavier Del Toro Garc

ıa

, Mar

ıa Jos

e Santoﬁmia Romero

elix Jes

us Villanueva Molina

and Juan Carlos L

opez L

opez

School of Computer Science, Computer Architecture and Networking Group, University of Castilla-La Mancha,

Paseo de la Universidad N

4, Ciudad Real, Spain

Keywords:

Affective Computing, Learning, Stress, Galvanic Skin Response, Eye-Tracking, Facial Expression, Electroen-

cephalography.

Abstract:

Detecting emotions in a learning environment can make the student-learning process more efﬁcient, avoiding

stressful situations that might eventually lead to failure, frustation and demotivation. The work presented here

describes a perceptive desktop devised to capture the sensations of any person facing learning activities. To

this end, we propose a perceptive environment enhanced with capabilities to perform an analysis of electroen-

cephalography, facial expression, eye tracking and particularly a very distinctive indicator of stress as it is

the galvanic response of the skin. This work focuses on the galvanic response of the skin, comparing the

performance of two devices in the context of the perceptive desktop. One of the devices was very attractive to

our environment as it was a mouse that ﬁt very well to our computer-based desktop, equipped with low-cost

sensors to detect the galvanic response. The other device is more tedious to place and more expensive but we

use it as a reference to know if the mouse is accurate. Four people were exposed to an experiment with the

two devices connected, and observing the results it can be concluded that there is no correlation between the

captures of both devices. Therefore, we could not select the mouse for our environment even though at ﬁrst it

looks like a very promising device.

1 INTRODUCTION

The use of student-facing learning analytics is gain-

ing attention as it provides information that can help

improving the teaching and learning process (Clow,

2012). Learning analytics could be correlated with

indicators derived from the student biosignal monitor-

ing. This information will eventually provide a more

accurate knowledge of the learning approach that bet-

ter ﬁts every student individual

The term affective computing is used to refer to

https://orcid.org/0000-0003-4700-5557

https://orcid.org/0000-0003-2893-0392

https://orcid.org/0000-0003-0343-8596

https://orcid.org/0000-0002-4177-4526

https://orcid.org/0000-0002-9996-8008

https://orcid.org/0000-0001-6132-1199

https://orcid.org/0000-0003-0773-5846

https://orcid.org/0000-0002-7372-1568

technological solutions that capture information about

a user’s emotions (Picard, 1995). In this sense, the

objective of this research is to analyze the feasibility

of a perceptive enhanced desktop, with capabilities to

assess students emotions. More speciﬁcally, the fo-

cus will be on identifying those symptoms related to

better performance during the learning process.

The proposed desktop is comprised of a set of de-

vices enabling the capturing of different signals. The

analysis of these signals will lead to an understanding

of the situations that generate discomfort or stress as

well as the identiﬁcation of the environmental factors

that impact the most on the student learning process.

In the ﬁrst part of this article we detail each of the

analyses that we are going to carry out in our percep-

tive desktop, brieﬂy describing the devices that we are

going to use to capture each emotion. At the end of

this section we will analyze Galvanic Skin Response

(GSR) from the point of view of two devices to be as

accurate as possible in capturing this parameter, as it

Chaparro, J., Navarro, R., Ruiz, A., Ruiz, J., García, X., Romero, M., Molina, F. and López, J.

An Affective-computing Approach to Provide Enhanced Learning Analytics.

DOI: 10.5220/0009368401630170

In Proceedings of the 12th International Conference on Computer Supported Education (CSEDU 2020) - Volume 1, pages 163-170

ISBN: 978-989-758-417-6

163

gives us valuable information on stress points. Later,

the results of the comparison made between the two

devices will be analysed, as well as the placement of

each one inside the perceptive desk. Finally, decisions

will be made as a result of the results obtained to ﬁnd

the precision in capturing emotions from our desktop

and that our future work through this environment to

provide guidance.

2 METHODS

Previous studies have used affective computing to

capture emotions and thus improve the user experi-

ence. In this way, some ﬁndings aim to increase the

motivation and engagement of users in learning tasks,

detecting their emotions within a multimodal environ-

ment (Garc

ıa-Garc

ıa et al., 2018). However, as tech-

nological solutions are currently in full development,

there are many challenges to be faced such as the

identiﬁcation of the state of emotions or the integra-

tion of information in multimodal environments (Wu

et al., 2016). We plan to extend the affective comput-

ing paradigm to promote learning activities adapted

to the student individual. To this end, we have de-

signed a perceptive desktop equipped with different

devices that, whenever possible will be unobtrusive

and non-invasive. First, the user will be prompted

to tailored stimuli, provoking them speciﬁc emotions.

This will be used to train the system on identifying

speciﬁc emotions. Then, the system will be able to

identify those emotions for which it has been trained

for and, when possible, will provide the teacher and

student the appropriate means to handle the situations.

One of the most signiﬁcant features to identify

stress and cognitive load is the GSR (Sriramprakash

et al., 2017) , but in other cases is studied with

some other factors such as speech to detect chronic

stress (Kurniawan et al., 2013). A desktop environ-

ment intended to be aware of stress symptoms has to

therefore be equipped with at least a device enabling

the GSR measurement. The analysis of the commer-

cial solutions brought about a mouse equipped with

the appropriate sensors to measure this feature. There

were very attractive features for this device, as it was

mainly its low-cost and unobtrusiveness. This de-

vice is initially devoted to gamers who want to track

their constants during a game. For this reason, we de-

cided to evaluate the performance of this mouse with

another device speciﬁcally devoted to measure GSR.

This other device will work as the gold standard, with

a more tedious setup but higher precision. The ﬁrst

step in designing the perceptive desktop is to evaluate

whether the performance of the mouse is enough to

evaluate stress symptoms.

The GSR provides relevant information about our

subconscious. Our skin reacts to different stimuli by

sweating and we can determine the conductance (or

resistance that is the inverse) of the skin at that mo-

ment. When the conductance is high, the skin sweats,

so we will dictate that the stimulus has provoked a

high activation or excitement. In other words, it indi-

cates a high emotional charge, although it is unknown

whether the effect is positive (happiness, excitement)

or negative (stress, fear, worry). Because of this, al-

though GSR is an ideal measure to track emotional

activation, it does not provide information about the

emotional valence, i.e. the quality of emotions.

In an effort to capture all the emotional data we

have designed a perceptual desktop depicted in Fig. 1.

This desktop is comprised of: 1) an electroencephalo-

gram headset to capture the electrical impulses of the

brain (Electroencephalogram (EEG)); 2) a depth cam-

era located above the monitor that will be used to de-

tect facial expressions; 3) an eye-tracking camera, lo-

cated in the lower frame of the monitor that will allow

us to determine where our user’s attention is focused;

4) the two aforementioned devices to measure GSR

and also the Electrocardiogram (ECG).

Figure 1: Design of a Perceptive Desktop with Devices to

Collect Data from Facial Expression, Eye-Tracking, Elec-

troencephalography, Galvanic Skin Response and Electro-

cardiogram.

The following subsections describe, in detail, the

main characteristics of the device and how they will

be used in order to detect emotions and, more specif-

ically, symptoms associated to stress.

2.1 Electroencephalography Analysis

The Brain Computer Interfaces (BCI) is a technol-

ogy based on obtaining the electrical activity of the

brain in order to control an external component. This

type of interface allows us to transform our thoughts

CSEDU 2020 - 12th International Conference on Computer Supported Education

164

into real actions, and its ﬁeld of application is very

wide-ranging: from the medical ﬁeld for the control

of robotic prostheses (Mcfarland and Wolpaw, 2010),

to the ﬁeld of leisure and video games (L

ecuyer et al.,

2008).

There are many devices created by different re-

search groups or manufactured by different compa-

nies. However, there are little differences among them

as they all have a similar basic function: measure

brain activity using sensors, process the signal obtain-

ing its most important characteristics and even inter-

act with the environment as desired by the user.

Despite the many technologies for the acquisition

of brain activity, the most effective tool nowadays is

the EEG due to its price, ﬂexibility and response time

for device control.

The Emotiv EPOC helmet is a non-invasive type

of BCI, i.e. its use does not involve physically dam-

aging or penetrating the skin or scalp. A kit including

the following items is provided with the product:

• An Emotiv EPOC headset (rechargeable lithium

battery).

• A Universal Serial Bus (USB) receiver.

• A hydration pack with 16 sensors.

• A saline solution.

• A USB charger.

The Emotiv EPOC headset has 14 channels that

are distributed on the basis of the international 10-20

system. The positioning of the channels is achieved

thanks to the plastic branches and two references that

ensure the correct position of the electrodes.

Figure 2: Emotiv EPOC Electrode Helmet for the Study of

Brain Electrical Activity or Electroencephalography.

The electrodes used by Emotiv EPOC (see Fig. 2)

have a small metal disc which is ﬁxed with a conduc-

tive paste, enabling very low contact resistance to the

signal. In addition to the electrodes, the headset has

a gyroscope and two accelerometers that provide in-

formation on the movements of the person’s head. It

also has a wireless transmitter that enables informa-

tion to be sent to the USB receiver and a battery that

powers it. Through these electrodes, treated with a

saline solution, we collect the information from each

sensor for later analysis and, most importantly, con-

trast this data with those collected by other sensors on

the desktop.

2.2 Facial Expression Analysis

Facial analysis aims to detect the muscle movements

that occur in the face, as it is the interface between the

world and the emotions. It is therefore the one that

allows us to have an image of what we feel, although

this can be provoked voluntarily or involuntarily.

There are several ways to retrieve data from fa-

cial expression. The Electromiography (EMG) is the

most accurate method of capturing data. However,

the desktop is equipped with a depth camera able to

capture facial movements and to identify and analyze

the user’s expression in a non-intrusive way. Depth

cameras basically work on the basis of two elements,

an infrared light projector and an infrared light sen-

sor. In this way, the camera projects beams of infrared

light, which the camera sensor will detect, being able

to collect the distance to each point and generate a

projection of what the camera captures.

Figure 3: RealSense D435 Depth Camera for the Study

of Facial Expression Located on the Top of the Perceptive

Desktop Monitor.

There are multiple depth cameras available on

the market for which different facial recognition al-

gorithms have been implemented. RealSense (see

Fig. 3) is one of them, enabling the acquisition of

depth images up to 90 fps with a resolution of up

to 1920x720. In order to properly capture the data,

the camera’s location and placement, the room’s light-

ing and the face’s visibility have to be taken into ac-

count. The information collected from the facial re-

sponse will indicate the validity of the emotional re-

sponse and, as with other devices on the desktop, we

An Affective-computing Approach to Provide Enhanced Learning Analytics

165

can compare the results with information obtained by

other sensors on the desktop.

2.3 Eye Tracking Analysis

The aim of the eye-tracking analysis is to establish

a relationship between the movements of the eye

and human cognition. Many ﬁelds are involved in

eye-tracking to analyze behaviors, and even people’s

trends. Some of these ﬁelds are: neuroscience, ad-

vertising, simulation, website testing, learning and

education, gaming, medical research. . . This technol-

ogy is also used for interacting with a computer sys-

tem without having to use the traditional keyboard or

mouse.

The dilatation of the pupil, the distance to the

screen, the direction of our gaze or eye blinking can

give us a lot of information about our visual attention,

concentration or even the things that we give more im-

portance. The operation of a non-intrusive eye tracker

is described by the Pupil Centre Corneal Reﬂection

(PCCR) technique, which basically consists in send-

ing a light to illuminate the eyes, so that the cameras

of the device collect those reﬂections, and through the

use of algorithms, the device can identify where you

are looking. It is important to have good lighting con-

ditions and a good positioning of the device when col-

lecting data.

Figure 4: Eye Tracking Tobii 4c to Do the Eye-Tracking

Study Located in the Lower Frame of the Perceptive Desk-

top Monitor.

Fig. 4 shows the Tobii 4c eye tracker, which will

enable the capture of eye tracking data. These data

will be fused with the rest of the data collected from

the other devices on the desktop.

2.4 Galvanic Skin Response Analysis

The variation of human sweat can give us enough in-

formation about how emotionally activated we are. In

order to analyse this, we have to look at the GSR, also

known as Electrodermal Activity (EDA) or Skin Con-

ductance Level (SCL). The interesting part of mea-

suring this parameter is that the sweat generated is

not consciously produced, but the nervous system ac-

tivates the sweat glands in the face of an alert. The

sympathetic nervous system is in charge of producing

this response.

The way to collect the data from SCL is basically

to place two electrodes on two ﬁngers, on the sole of

the foot or on the sole of the hand. The sensors are

usually silver/silver chloride and a constant voltage

(low level) is applied to them to measure the voltage

difference between the two electrodes. With this non-

invasive process you can collect emotional responses

that are useful for countless domains, such as psy-

chology, psychotherapy, marketing or even usability

studies.

The design used in this research will provide a

comparative analysis between two devices that mea-

sure the GSR with the aim of being able to have these

data available and make a further analysis of the stress

peaks. First, we will analyze the data collected with

Shimmer3 GSR+ Unit, then we will see in detail how

the data is collected with the Mionix mouse. This

comparison aims to evaluate different characteristics

or parameters of the devices used in order to select

one or the other depending on what is needed. Fi-

nally, the methodology to collect data from both de-

vices will be discussed in order to compare both sys-

tems as well as the assembly of the different sensors in

what will be the perceptual desktop designed to cap-

ture emotions.

2.4.1 Accuracy Evaluation using Shimmer3

GSR+ Unit

Shimmer is an Irish company that was originally con-

ceived in 2006, although it was not founded until

2008. Since then, it has been regarded as a pioneer

in portable sensor technology and solutions.

One of its products is the GSR Unit+, which

measures galvanic skin response (GSR) through two

reusable electrodes placed on two ﬁngers of one hand.

In response to internal and external stimuli, sweat ap-

pears on our skin that makes the glands more active,

increasing the moisture content in the skin and al-

lowing the electrical current to ﬂow more easily by

changing the balance of positive and negative ions in

the secreted ﬂuid. In this way, the conductance of the

skin increases and the resistance decreases.

To gather the measured data the Consensys pro-

gram, provided by Shimmer company, is employed,

following the next steps:

CSEDU 2020 - 12th International Conference on Computer Supported Education

166

Figure 5: Shimmer3 GSR for the Capture of Galvanic Re-

sponse Signals from the Skin, Located in the Proximal Pha-

lanx of the Left Hand.

Table 1: Shimmer3 GSR+ Unit Characteristics.

Current draw 60 µA

Measurement range

8kΩ - 4.7MΩ

(125µS - 0.2µS)

+/- 10%,

22kΩ - 680kΩ

(45µS - 1.5µS)

+/- 3%

Frequency range DC-15.9Hz

Input protection

RF/EMI ﬁltering;

Current

limiting; GSR

inputs include

deﬁbrillation

protection.

Connections

GSR 1 input (red),

GSR 2 input

(black):

Hospital grade

1mm

contact proof

IEC/EN

60601-1 jacks

DIN42-802.

Analog/digital

auxiliary

input: 4 position

3.5 mm connector.

Bias voltage

across GSR inputs

0.5V

EEPROM memory 2048 bytes

Weight 30g

1. With the device turned on and plugged into the

dock by USB, we start the Consensys program.

2. We select our Shimmer and put the ﬁrmware that

interests us, in this case we record the LogAnd-

StreamShimmer v0.11.0 to obtain data by blue-

tooth and to obtain information of registry in the

SD card.

3. The next step is the conﬁguration to measure what

we need. For example we can change the data

record, the name of our test and device, the sam-

pling frequency and what we want to measure. At

the end we have to write the conﬁguration in the

device.

4. Place the sensors before collecting the data and

then place them where they belong.

5. The last step is to go to live data, connect the blue-

tooth and look at the data we are getting.

Fig. 5 depicts the appropriate emplacement of the

electrodes. In this case, they are placed on the index

and middle ﬁngers in the proximal phalanges. After

pressing the orange button the device will start cap-

turing data. The last step would be to download and

visualize the data, and for it, the Consensys program

can be used to download the data in a ﬁle Comma-

Separated Values (CSV) so we can analyze the GSR.

2.4.2 Accuracy Evaluation using Mionix QG

Mionix is a Swedish company that is known for the

development and sale of high performance gaming

accessories designed for perfect ergonomics. Their

products include keyboards, headphones, and mouse.

The product that we will use in our perceptive en-

vironment is the Mouse Mionix Naos QG which has

additional sensors that will allow us to monitor heart

rate, GSR and the performed activity.

If we look at the sides of the mouse logo, we ﬁnd

on the left the optical sensor for measuring heart rate,

and on the right two small metal connectors to get

the response of the skin. The operation of the opti-

cal sensor is similar to the operation of sensors found

in smartwatches, on the other hand, the galvanic skin

response sensor measures the conductivity of the skin

detecting the level of stress, because the more stress

the user has, the more sweat in the palm of the hand

and therefore more conductivity.

It can be seen that the features, design and the

market to which Mionix is directed are designed for

gamers or players to play in a comfortable, accurate

way and in this particular model is obtained data on

how the player is reacting during the game. However,

in the design of our perceptive desktop we thought

about this mouse to identify activities in which the

user is comfortable or not.

The software that has the mouse is called Mionix

Hub, which can be downloaded for free for Windows

and MAC to monitor what the sensors read at every

moment of the activity. However, there is a Web-

socket Application Programming Interface (API) for

the data to be accessible from any environment and

An Affective-computing Approach to Provide Enhanced Learning Analytics

167

Figure 6: Mionix Naos QG Mouse for the Capture of Gal-

vanic Response Signals from the Skin Located in the Right

Hand.

for any activity for which we need this data(Wulff-

Abramsson, 2017).

A Unity program has been developed for data ac-

quisition using this API. This program allows us to

collect the information from the GSR in a text ﬁle

with JavaScript Object Notation (JSON) format.

2.4.3 Experiment Design

The experiment was conducted with 4 volunteers (2

men and 2 women). We have chosen a game that

keeps the user stressed during the ﬁve or ten min-

utes of the test, in order to see sudden changes in

the galvanic response of the skin, and thus be able

to more clearly identify and compare the skin con-

ductance levels in both devices. The test consisted of

playing slither.io(Howse, 2016), which is a massive

online game involving multiple users which is aimed

at growing your worm or snake through the food on

the map or other worms or snakes that have been elim-

inated by hitting their head on the body of another

worm or snake. The game, therefore, ends when you

hit another worm or snake.

In order to collect data on the galvanic response

of the skin, an environment with these components is

proposed:

• GSR with Shimmer3: Connected to the computer

via Bluetooth for data collection, and placed on

the left hand of the user who performs the test.

The placement of the device is carried out as ex-

plained in the sub-section 2.4.1.

• GSR with Mionix QG: Connected to the computer

via USB the user will handle it with the right hand

placing the palm of the hand correctly on the elec-

trodes of the mouse.

• Camera for recording user’s expressions: The

camera is used to get user snapshots and be able

to compare their expressions with the data GSR.

• Screen snapshots: The screen is captured to iden-

tify the moment in which any alteration may occur

in the user.

Once the devices are ready, we run the program

to collect data with Shimmer (Consensys), and we

also run the program we developed in Unity, taking

into account that the sampling frequency of the two

devices are the same. For this experiment we have

decided to capture 1 sample every second (1Hz), be-

cause according to (Geddes and Baker, 1991) the low

frequencies are ideal for obtaining the GSR (a range

of 0 Hz to 5 Hz is recommended for the collection of

this type of measures).

The test takes between 5-10 minutes, and all data

and visuals are saved for further analysis. It should be

noted that the videos will only be used to make a com-

parison with the data collected from the GSR devices

and therefore will be nonpublished material. Never-

theless, the test participants signed a consent form that

makes it clear that their images will not be published

at any time, and that they will only be used for com-

parative testing in this study.

3 RESULTS

3.1 Shimmer3 vs. Mionix Naos QG

When the experiment was over, we could observe the

different characteristics of the two devices. First, the

range in which the Shimmer3 conductance values are

found is between 125µS and 0.2µS as shown in Tab. 1.

However, the values we found in the Mionix measure-

ment range are between 0µS and 1µS. Measurement

ranges did not match even though we set Shimmer3

to range 3 (Tab. 2) which places values between 1.5µS

and 0.2µS.

Table 2: Scale Conductance Range Shimmer3.

Range 0 125µS to 15.9µS

Range 1 15.9µS to 4.5µS

Range 2 4.5µS to 1.5µS

Range 3 1.5µS to 0.2µS

Range 4 Auto-Range

On the other hand, we found that when correlating

the data obtained in each device, there is no linear

relationship in any of the users exposed to the test. As

can be seen in Tab. 3 the results obtained are weak

negative (User 1, User 2, User 4) and positive (User

3) relationships.

For a clearer view of the comparison, we can see

in Fig. 7 and Fig. 8 that the collected data have no

CSEDU 2020 - 12th International Conference on Computer Supported Education

168

correspondence, and furthermore, in Fig. 8, we can

clearly see the moments in which user 1 had higher

values of conductance, however in Fig. 7 we are not

able to detect the moments in which the user has some

kind of excitation.

Figure 7: Results Collected on User 1 Using the Mionix

Naox QG Mouse.

Figure 8: Results Collected on User 1 Using the Shimmer3

GSR Device.

3.2 Perceptive Desktop

The integration of all the devices described for the de-

velopment of the perceptive desktop can be seen in

Fig. 9.

Figure 9: Final Appearance of the Perceptive Desktop with

All Devices for Capturing Data.

First, the depth camera is placed on top of the

monitor at an ideal angle for collecting facial expres-

sions. Second, the eye tracker is placed in the lower

frame of the monitor for eye tracking. On the other

hand, when the user is seated on the desktop we will

place him/her the EEG headset, the shimmer for the

detection of the GSR and ECG and the mouse will be

used by the user to carry out the different experiments.

Table 3: Correlation Coefﬁcient.

User 1 -0.026811939

User 2 -0.083538697

User 3 0.066392937

User 4 -0.008630607

4 DISCUSSION

According to the obtained results in the comparison

of the two GSR devices, we have observed that the

mouse reveals limitations when it comes to perceive

changes in GSR. The ﬁrst fact that we observed to

determine the limitations that the mouse had with re-

spect to the other device was that the range of val-

ues in which the mouse was moved was a range of

values that did not correspond to the values obtained

with the Shimmer3 GSR. To determine this ﬁrst limi-

tation we based on the fact that typical skin resistance

values vary from 47 kΩ to 1 MΩ (21µS to 1µS con-

ductivity) (ShimmerSensing, 2015). This fact, made

us repeat the tests on the same users to be sure that

everything was being carried out correctly, since for

example readings may be altered by hand movements

during the test or be affected by the position of the

palm on the mouse. This leads to a lack of consistency

on the observed reading during the testing phase.

It is observed that even in Shimmer, when there is

some movement in the electrodes there is ﬂuctuations

in data that could not be taken as good. For the assem-

bly of the other devices of the desktop, it is necessary

to bear in mind that they are well placed to collect the

data, but also the different programs that collect the

data have the same sampling frequency for accurate

analysis.

In our quest to get an explanation of the values we

were getting from the mouse, we sent an email to the

Mionix Support Portal to ﬁnd out if what we were get-

ting was in other units or if the data we were getting

was from some conversion. The only answer we re-

ceived was that the units should be in microsiemens.

Finally, we decided not to take any more samples

in the experiment because we saw that the mouse gave

us unreliable results, even though we repeated the test

over and over again or even collected data from dif-

ferent users.

An Affective-computing Approach to Provide Enhanced Learning Analytics

169

5 CONCLUSION

The use of learning analytics is gaining attention as it

support teachers towards improving the learning pro-

cess of each student, by shaping the learning process

to the approach that better ﬁts the student cognition

process. This work presents a desktop environment

setup, enhanced with a set of devices capable of per-

ceiving the student emotions, with special focus on

identifying the symptoms associated to stress.

This work presents the desktop devices along with

the purposes to which each device is intended. Special

attention has been devoted to the Galvanic Skin Re-

sponse (GSR) as it is a reliable sign to detect changes

on emotions. Two devices have been evaluated and

compared in order to capture GSR data. One is based

on mouse-like device which, a priori, looks like a

very attractive device for being an element present in

every computer-based desktop. The other implies a

more tedious setup, with electrodes tied to two ﬁn-

gers.

An experiment was conducted to evaluate the ac-

curacy and performance of the mouse-based device

(the Mionix) using the Shimmer device as the gold

standard. Results have demonstrated that despite be-

ing an attractive device for the construction of a per-

ceptive desktop, the obtained measures are not reli-

able enough for the sought purpose. Further works

will consist in combining the data obtained from the

different devices and obtain a consistent pattern of

work-related stress singnals of individuals.

ACKNOWLEDGEMENTS

This work was supported by European Union’s Hori-

zon 2020 programme under grant agreement ID

857159, project SHAPES (Smart and Healthy Age-

ing through People Engaging in Supportive Sys-

tems); Spanish Ministry of Science, Innovation

and Universities under Grant PLATINO (TEC2017-

86722-C4-4-R); Spanish Ministry of Economy and

Competitiveness under Grant CITISIM (TSI-102107-

2016-8 ITEA3 N

15018); Regional Government

of Castilla-La Mancha under Grant SymbIoT (SB-

PLY/17/180501/000334); Spanish Ministry of Edu-

cation, Culture and Sport under Grant FPU Program

(ref. FPU 16/06205); and Xavier del Toro Garc

ıa

has received ﬁnancial support from the European Re-

gional Development Fund (Fondo Europeo de Desar-

rollo Regional, FEDER).

REFERENCES

Clow, D. (2012). The learning analytics cycle: Closing the

loop effectively. In Proceedings of the 2Nd Interna-

tional Conference on Learning Analytics and Knowl-

edge, LAK ’12, pages 134–138, New York, NY, USA.

ACM.

Garc

ıa-Garc

ıa, J. M., Penichet, V., Lozano, M., Garrido, J.,

and Law, E. (2018). Multimodal affective computing

to enhance the user experience of educational software

applications. Mobile Information Systems, 2018:1–10.

Geddes, L. A. and Baker, L. E. (1991). Principles of Ap-

plied Biomedical Instrumentation. Wiley, London, 3rd

edition.

Howse, S. (2016). Slither.io. urlhttp://slither.io/. Accessed

15-07-2019.

Kurniawan, H., Maslov, A., and Pechenizkiy, M. (2013).

Stress detection from speech and galvanic skin re-

sponse signals. In Proceedings of the 26th IEEE In-

ternational Symposium on Computer-Based Medical

Systems, pages 209–214.

ecuyer, A., Lotte, F., Reilly, R. B., Leeb, R., Hirose, M.,

and Slater, M. (2008). Brain-computer interfaces, vir-

tual reality, and videogames. Computer, 41(10):66–

72.

Mcfarland, D. J. and Wolpaw, J. R. (2010). Brain–computer

interfaces for the operation of robotic and prosthetic

devices. Advances in Computers, 79:169–187.

Picard, R. W. (1995). Affective computing. Technical re-

port, Massachusetts Institute of Technology.

ShimmerSensing (2015). GSR+ User Guide Revision 1.8.

Sriramprakash, S., Prasanna, V., and Murthy, O. (2017).

Stress detection in working people. Procedia Com-

puter Science, 115:359–366.

Wu, C., Huang, Y., and Hwang, J. (2016). Review of af-

fective computing in education/learning: Trends and

challenges. British Journal of Educational Technol-

ogy, 47:1304–1323.

Wulff-Abramsson, A. (2017). Communication framework

for the mionix naos qg mouse for online and ofﬂine

usage. Technical report, Augmented Cognition Lab at

Aalborg University Copenhagen.

CSEDU 2020 - 12th International Conference on Computer Supported Education

170