HealthyClassroom

A Proof-of-Concept Study for Discovering Students’ Daily Moods and Classroom

Emotions to Enhance a Learning-teaching Process using Heterogeneous Sensors

Minh-Son Dao

1, 2

, Duc-Tien Dang-Nguyen

, Asem Kasem

and Hung Tran-The

Universiti Teknologi Brunei, Brunei

University of Information Technology, Vietnam

Dublin City University, Ireland

National Institute of Information and Communications Technology, Japan

Keywords:

Moods, Emotions, Personal Lifestyle, Physiological Data, Wearable Sensors, Teaching and Learning Process,

Data Visualization.

Abstract:

This paper introduces an interactive system that discovers students’ daily moods and classroom emotions to

enhance the teaching and learning process using heterogeneous sensors. The system is designed to enable

(1) detecting students daily moods and classroom emotions using physiological, physical activities, and event

tags data coming from wristband sensors and smart-phones, (2) discovering association/correlation between

students’ lifestyle and daily moods, and (3) displaying statistical reports and the distribution of daily moods

and classroom emotions of students, both in individual and group modes. A pilot proof-of-concept study was

carried out using Empatica E4 wristband sensors and Android smart-phones, and preliminary evaluation and

ﬁndings showing promising results are reported and discussed.

1 INTRODUCTION

The inﬂuence of moods/emotions on education has

been researched for decades. Many researches try to

understand the impact of moods/emotions on different

aspects of education in order to create a better lear-

ning environment for students (Pekrun, 1992)(Bryan

et al., 1996)(Titsworth, 2010).

Generally, there are four approaches to under-

stand the impact of moods/emotions in learning and

working environments: (1) paper-based questionnai-

res (Pekrun, 1992)(Bryan et al., 1996)(Lewine et

al., 2015)(Liew and Tan, 2016), (2) digital-based

event tags (Rachuri et al., 2010)(Sottilare and Proctor,

2012)(Kikamwa et al., 2013)(Gjoreski et al., 2015),

(3) physiological signals (Wioleta, 2013)(Hang et al.,

2014)(Valenza et al., 2014), and (4) hybrid (Sano and

Picard, 2013)(Zenonos et al., 2016).

The ﬁrst approach mostly uses survey methods to

get samples. This leads to a small volume of sam-

ples, limitations in tracking mood/emotion states, sta-

tic and inﬂexible way to collect data, as well as bias

and delay in providing answers. These factors in-

ﬂuence expected outcomes and make it hard to ge-

neralize the research into higher conceptual levels,

scale up into larger scopes, as well as enabling real-

time monitoring and prediction/detection of moods

and emotions of learners.

The second approach uses computers or smart-

phones as a portal to collect data from users for de-

tecting and/or predicting mood and emotion states.

It is common in the methods of this approach to de-

sign an application (or a website) accessed by smart-

phones or computers with predeﬁned mood/emotion

categories and levels. Users interact with this appli-

cation to input related information for detection of

moods and emotions. Where relevant, prediction mo-

dels could be built using collected data to allow au-

tomatic prediction of users moods/emotions. Nevert-

heless, this approach cannot prevent bias of answers

since users sometimes ﬁnd it difﬁcult or inconvenient

to determine exactly how they feel at any given time.

The third approach leverages on physiological sig-

nals collected from users using wearable sensors.

Physiological data could be a good alternative to con-

veniently capture information from humans, and to

objectively detect moods/emotions. This is because

such data directly reﬂects, at least as far as the cho-

sen signal sources are concerned, how human bo-

dies react in a surrounding environment with respect

Dao, M-s., Dang-Nguyen, D-T., Kasem, A. and Tran-The, H.

HealthyClassroom - A Proof-of-Concept Study for Discovering Students’ Daily Moods and Classroom Emotions to Enhance a Learning-teaching Process using Heterogeneous Sensors.

DOI: 10.5220/0006749106850691

In Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2018), pages 685-691

ISBN: 978-989-758-276-9

685

to any faced issues (Wioleta, 2013). The wearable

sensors used to collect such physiological data vary

from uncomfortable bio-sensors (Hang et al., 2014),

comfortable t-shirts with integrated fabric electrodes

and sensors (Valenza et al., 2014), to wristband sen-

sors (Thapliyal et al., 2017)(Koskimaki et al, 2017).

Among these types of sensors, wristband sensors are

known to be more comfortable in terms of mobility,

mounting location, and continuous skin connection.

The last approach combines smart-phones and we-

arable sensors to collect physiological data, human

behaviour, and moods/emotions tags from users as

well as displaying the detected results to users via

smart-phones (Sano and Picard, 2013)(Zenonos et al.,

2016). This approach leverages all advantages of

smart-phones and wearable sensors to collect data and

detect moods/emotions while still maintaining high

level of convenience and freedom for users.

From psychological perspective, moods and emo-

tions are used to express feelings that people expe-

rience. Although these words are frequently used

interchangeably, there are many differences between

moods and emotions (Beedie et al., 2005). The main

ones are related to intensity, the objects people react

to, and the duration they last. While emotions are

intense feelings directed at someone or something,

moods are feelings that tend to be less intense and

often (though not always) lack a contextual stimu-

lus (Robbins et al., 2010). Besides, a mood may last

longer than an emotion (e.g. hours/days versus se-

conds/minutes) (Beedie et al., 2005).

In order to avoid any misunderstanding, in this pa-

per we use emotions for expressing the feelings of stu-

dents in a classroom where students (university level

in current study) react to a lecturer, a lesson activity,

or their classmates within a limited time; and moods

for out-of-classroom feelings that are heavily inﬂuen-

ced by the environment, physiology, or mental state.

In other words, the contexts of emotions and moods

are within-a-classroom and throughout-the-day, re-

spectively.

Many researches have tried to understand the im-

pact of moods and emotions on education based

on two independent contexts: in classroom (Hang

et al., 2014)(Lewine et al., 2015)(Liew and Tan,

2016)(Mainhard et al., 2017), and on campus (Bryan

et al., 1996)(Febrilia and Warokka, 2011)(Kikamwa

et al., 2013)(Gjoreski et al., 2015). Although these

studies have found useful results, there are still open

questions that need to be investigated thoroughly:

1. The research carried out by (Mainhard et al.,

2017) highlights that the students’ emotional ex-

periences can be driven by evolving the speciﬁc

relationship between lecturers and students during

class time. Therefore, it would be very useful

to have a tool that can visualize students’ emo-

tions during a class to allow a lecturer to ﬂexibly

change his/her educational method or activity in

order to stimulate positive emotions that enhance

the teaching-learning process.

2. There seem to be no work we are aware if on

discovering association/correlation between daily

moods and classroom emotions of learners. For

example, if one student comes to a class with ne-

gative moods (e.g. nervous, stressed) due to not

yet ﬁnishing his/her assignment though staying

late last night, whether this student can have ex-

cited and relaxed emotions during the class time

though a classroom’s atmosphere is very exciting?

3. There are few studies that investigated associa-

tion/correlation between a student’s lifestyle and

his/her daily moods. In this study, we are inte-

rested to investigate, for example, if long-term ne-

gative moods of a student are correlated with a

certain lifestyle (e.g. staying up late, exercising

too much, lack of physical activity, etc.). This

may enable student’s counseling to be triggered

automatically and provides valuable input for the

counseling process.

This paper addresses the above questions at a uni-

versity level by creating a framework, namely He-

althyClassroom, which uses wristband sensors and

smart-phones to collect physiological, physical acti-

vities, and event tags data to detect and visualize

students’ daily moods and classroom emotions, as

well as to discover any association/correlation bet-

ween students’ lifestyles and daily moods.

2 HEALTHYCLASSROOM

Figure 1 illustrates an overview of the Healthy-

Classroom framework. In this context, mood and

emotion can be used interchangeably according to

the context (i.e. if the framework is applied inside a

classroom, emotion is considered, otherwise mood

is used). The physiological, location, and event

tags data (when users provide input by tagging

speciﬁc events) are sent from wristband sensors

and smart-phones to a cloud-computing platform.

Using this data, students moods and emotions

are then detected/recognized and displayed on a

website that can be accessed by lecturers and/or

administration users. The timeliness of such in-

formation provides valuable input to understand

the ongoing teaching-learning process and improve

the delivery and interaction between lecturers and

INDEED 2018 - Special Session on INsights DiscovEry from LifElog Data

686

Figure 1: The HealthyClassroom Overview.

students. The data generated from activities outside

the classroom can be used not only to directly inﬂu-

ence the learning process, but also to understand the

social daily activities of students and their mental he-

alth, in addition to triggering necessary counseling in-

tervention.

The system works as follows: The wristband de-

vices worn by students frequently send physiologi-

cal and location signals to smart-phones via Blue-

tooth protocol. The smart-phones do preprocessing

(i.e. data ﬁltering and normalizing) and capture tag-

ging data from students (if provided). For example,

students may choose at different times to give feed-

back about the current mood/emotion they feel, or the

physical activity they are doing. This is performed

using a smart-phone application synchronized with

the wristband sensor, and essentially helps in gene-

rating labeled data crucial at the starting phase of our

study. The data is then sent to a cloud-computing sy-

stem for storage and analysis. In the cloud, training

and re-training tasks to create classiﬁers for each type

of mood/emotion are triggered according to predeﬁ-

ned models and thresholds. These classiﬁers will de-

tect mood/emotion along a time-axis. The results can

be displayed - in different forms - on a website that

can be accessed by concerned users for further acti-

ons such as re-arranging students classroom positions,

re-adjusting lessons content, activating classroom re-

sponse system, counseling identiﬁcation for students

who exhibit negative moods and emotions for a long

time, etc.

2.1 Models of Moods/Emotions

In our work, we refer to (Wioleta, 2013) for a survey

on emotion’s models based on various theories of

emotions and affect models developed for different

application ﬁelds. The conclusion is that creating a

Figure 2: The structure of mood, reused from (Robbins et

al., 2010).

model of emotion is not a trivial problem. In this rese-

arch, we use the affect structure presented in (Robbins

et al., 2010), and illustrated in Figure 2.

2.2 Affect Detection and Lifestyle-Mood

Pattern Discovery

Figure 3 illustrates how the association/correlation

among lifestyle, daily moods, classroom emotions,

and classroom interaction can be taken into account

and visualized properly. There are four processes

run synchronously: (1) physical activities (PA) de-

tection using accelerometer and gyroscope data ex-

tracted from smart-phones, (2) daily moods (DM)

recognition using physiological data collected from

wristband sensors, (3) classroom emotions (CE) dis-

covery using physiological data collected from wrist-

band sensors, and (4) association/correlation uncove-

ring using the outputs of the three previous processes.

The ﬁrst process is built by utilizing the PHA-

SOR system (Dao et al., 2017) where accelerometer

and gyroscope collected from smart-phones are trans-

lated into a set of human daily activities, namely stan-

ding, walking, jogging, up-stairs, downstairs, and sit-

ting. We have also added one more activity, driving,

to this set.

The second and third processes are constructed

based on the concepts shown in Figure 1. Figure 4

illustrates the mechanism used to detect moods and

emotions using physiological signals and event tags.

The fourth process leverages the research presen-

ted in (Laleh et al., 2015) to discover lifestyle-mood

patterns (i.e. association/correlation between lifestyle

and daily moods of students). In this case, we as-

sume that human daily activities could have associ-

ation/correlation with moods. In other words, a se-

quence of activities could lead to a certain mood sta-

tus during a period of time. We formalize this associ-

ation as follows. Let E

denote an activity at time t, ∆t

HealthyClassroom - A Proof-of-Concept Study for Discovering Students’ Daily Moods and Classroom Emotions to Enhance a

Learning-teaching Process using Heterogeneous Sensors

687

Figure 3: The personal dashboard of associations/correlations among lifestyle, daily moods, and classroom emotions of a

student visualized in HealthyClassroom.

denote the ”time lag” between two consecutive activi-

ties E

and E

i+1

, and F

represent a mood’s label. The

association/correlation between n consecutive activi-

ties and mood status F can be expressed by the pattern

ρ = (F

: E

;∆t

;...;∆t

n−1

: F

) (1)

where F

and F

is the previous and current mood sta-

tuses. Then, the pattern recognition component em-

ploys an extended ﬁnite-state automata (FSA). The

automaton is designed to contain a ﬁnite number of

states and state transitions. The FSA is a 5-tuple,

(OS, T S, E, F

, F

), consisting of a ﬁnite set of ordi-

nary states (OS), time states (TS), directed edges (E),

a start state F

∈ OS and a ﬁnal or acceptance state

∈ OS. The ”Algorithm 1” described in (Laleh et al.,

2015) is utilized to count the frequencies of patterns.

Figure 4: The Mood/Emotion Detection Mechanism.

If a pattern occurs frequently, it might be signiﬁcant

and could be, in our context, argued as a lifestyle-

mood pattern.

3 EXPERIMENTAL RESULTS

In this section, we describe HealthyClassroom’s wor-

king environment and performance evaluation. Cur-

rently, the HealthyClassroom works as a proof-

of-concept framework for discovering students’

daily moods and classroom emotions to enhance a

teaching-learning process. Therefore, only the three

questions mentioned in Section 1, as a signiﬁcant con-

tribution of the HealthyClassroom, are considered for

evaluation.

3.1 Working Environment Setting

We use the Empatica E4 wristband (E4)

as a weara-

ble sensor for our system.

The E4 is recently used to collect physiologi-

cal data for early detection of migraine attacks (Ko-

skimaki et al, 2017), and it is shown that this data is

good enough for doing experiments. The E4 has the

following sensors: (1) Photoplethysmograph (PPG)

sensor: measures blood volume pulse (BVP) from

https://www.empatica.com/e4-wristband

INDEED 2018 - Special Session on INsights DiscovEry from LifElog Data

688

which heart rate, heart rate variability (HRV), and ot-

her cardiovascular features may be derived; (2) 3-axis

Accelerometer (ACC): captures motion-based acti-

vity; (3) Electrodermal Activity (EDA) sensor: me-

asures sympathetic nervous system arousal and to de-

rive features related to stress, engagement, and exci-

tement; and (4) Infrared Thermopile (TEMP): reads

peripheral skin temperature. Besides, E4 also has a

function called Event Mark Button that is used to tag

events and associate them with physiological signals.

There are three working modes by which users can

collect their data: (1) Recording Mode: uses an inter-

nal memory to store data that can be copied via USB

port later; (2) Streaming Mode: connects to a smart-

phone or desktop computer via Bluetooth using Real-

time App and mobile API for desktop and mobile de-

vices, respectively; and (3) Upload to Connect: uses

Empatica’s secure cloud platform. Figure 5 displays

one example of data collected from E4.

We have developed a smart-phone app that gets

activated by the E4 Event Mark Button, which is used

to tag the events using the list of mood/emotion labels

mentioned in subsection 2.1. This app can run with

manual and semi-automatic options. The former gets

a label from the user, and the latter generates a label

by using the second and third processes described in

subsection 2.2. A user can re-tag an auto-generated

label if he/she does not agree with the label generated

by the system.

We have also built a website through which users

can login to visualize their information including phy-

sical activities (PA), daily moods (DM), and class-

room emotions (CM). If the user has a lecturer role,

he/she can visualize his/her students information in

real-time mode, to enable taking quick actions such

as adjusting lesson activities (maybe to achieve more

positive in-class emotions).

We have also implemented four system compo-

nents corresponding to the four processes mentioned

in subsection 2.2: (1) Physical Activities Detection,

(2) Mood Detection, (3) Emotion Detection, and (4)

Lifestyle-mood Pattern Discovery.

Figure 5: Physiological data collected by E4, recorded from

18:20 November 7, 2017 to 17:00 November 8, 2017.

3.2 Evaluation

We have asked students and lecturer of one univer-

sity course to join our pilot experiment (15 students, 1

lecturer, Introduction to Programming course). Each

student is requested to wear the E4 wristband for con-

tinuous three weeks time, except during the time the

take bath.

The ﬁrst week: we collect samples for training

mood/emotion detection models. Students reported

some inconvenience in continuously tagging events

(e.g. choosing mood/emotion labels), especially du-

ring class time since emotions change very fast. We

also use the data of this week to track the ratio of over-

all positive and negative emotions.

The second week: we automatically detect in-

class emotions, and ask the lecturer to monitor the

students emotions visualized by the system, and to

adjust teaching activity or classroom setting to im-

prove classroom emotional climate. The system sho-

wed that high positive emotions have increased by

20% compared to previous week. Currently, we do

not have strong evidence to attribute this improvement

to the adjustment alone, but we argue that knowing

students’ emotions is an important factor that helps to

improve classroom emotional climate. In this week as

well, we start to discover the lifestyle-mood patterns.

The third week: at the beginning of this week,

we released the lifestyle-mood patterns with the hig-

hest negative emotions for each student, and asked

them to try to change these patterns. For example, one

pattern was that staying up late was associated with

”stressed” emotion. Another pattern was that sitting

for long time in campus was associated with ”bored”

emotion. At the end of this week, we noticed an in-

crease of 30% for high positive emotions compared to

the ﬁrst week. Again, we do not have strong support

for the reason of this improvement, but we argue that

discovering the lifestyle-mood pattern could be anot-

her factor that improves classroom emotional climate.

It is important to note that from the second

week onwards, the system can automatically detect

moods/emotions and daily activities, and therefore the

events tagging has decreased (i.e. only for correcting

wrong labels). Therefore, students expressed feeling

more comfortable with the HealthyClassroom compa-

red to the ﬁrst week.

By surveying students feedback, ten out of ﬁfteen

students (66.7%) agreed that HealthyClassroom can

help them control their lifestyle-mood relation as well

as improve their passion in studying. Two students

(13.3%) felt nothing changed with the use of Healt-

hyClassroom. Three students (20%) had neutral com-

ments.

HealthyClassroom - A Proof-of-Concept Study for Discovering Students’ Daily Moods and Classroom Emotions to Enhance a

Learning-teaching Process using Heterogeneous Sensors

689

The lecturer felt excited when using Healthy-

Classroom. Nevertheless, continuously monitoring

the percentage of positive emotions and ﬁnding a way

to cheer up the class somehow distract the teaching

process. There should be a function that can alert the

lecturer when the emotion climate reaches a certain

level so that the lecturer can adjust himself/herself

better. This solution can alleviate the burden of moni-

toring the ﬂuctuation of classroom emotions continu-

ously. This matter is recognized and will be adopted

in future works.

Since the human daily activity, mood, and emo-

tion detectors are built by using PHASOR (Dao et al.,

2017), we have used the same evaluation scheme to

evaluate the ﬁrst three processes. According to our

evaluation, the ﬁrst process (i.e. activities detection)

reaches 90% accuracy while the second (i.e. moods

detection) and the third (i.e. emotions detection) gain

65.2% and 57.3%, respectively.

4 CONCLUSIONS

In this paper, we introduced the HealthyClassroom

framework that works with wristband sensors, smart-

phones, and cloud computing environment to analyze

the impact of students’ daily moods, classroom emo-

tions, and lifestyle on the teaching-learning process.

The HealthyClassroom provides a data visualization

for lecturers to track classroom emotions in real-time,

to enable adjusting lessons’ activities in order to im-

prove students positive emotions. Last but not le-

ast, by getting insights from recorded personal-lifelog

data, including lifestyle-moods patterns, students can

be counseled on how to improve their studying pro-

cess by changing their lifestyle activities.

Although the HealthyClassroom currently is a pi-

lot proof-of-concept study, important ﬁndings have

have been gained. Some interesting lifestyle-moods

patterns are discovered that can help lecturers and stu-

dents adjust their teaching-learning behaviors.

We recognize that our current experimental setup

is limited to a small uncontrolled study, and in the fu-

ture it is important to address this issue in order to

reliably conclude if the use of HealthyClassroom re-

ﬂects positively on the teaching-learning process. Be-

sides, it will be interesting to further analyze the cor-

relation between the emotions of different students,

and to see if a students mood/emotion could be in-

ﬂuenced by another student(s). To tackle such in-

vestigations, it might be important to monitor class-

room environment using some sort of crowd-sensing

(e.g. activating smart-phones’ microphones to mea-

sure noise levels, monitoring smart-phone-user inte-

raction to identify students’ distraction, indoor loca-

tion to detect possible students interactions, etc.).

ACKNOWLEDGMENTS

This work is supported by the Internal Research

Grant of Universiti Teknologi Brunei, grant number:

UTB/GSR/1/2016 (10). The authors would like to

thank the students and lecturers of Universiti Tekno-

logi Brunei for their voluntary participation in this

project.

REFERENCES

Beedie, C.J. and Terry, P.C. and Lane, A.M. (2005). Dis-

tinctions between emotion and mood. COGNITION

AND EMOTION. 19 (6), 847-878

Bryan, T. and Mathur, S. and Sullivan, K. (1996). The Im-

pact of Positive Mood on Learning. Learning Disabi-

lity Quarterly. 19 (3), 153-162

Dao, M.S. and Dang-Nguyen, D.T. and Riegler, M. and

Gurrin, C. (2017) Smart lifelogging: recognizing hu-

man activities using PHASOR. In Proc. of 6th Int.

Conf. on Pattern Recognition Applications and Met-

hods (ICPRAM 2017). 24-26.

Febrilia, I. and Warokka, A. (2011). The Effects of Positive

and Negative Mood on University Students Learning

and Academic Performance: Evidence from Indone-

sia. In Proc. of 3rd Int. Conf. on Humanities and So-

cial Sciences. 1-12

Gjoreski, M. and Gjoreski and H. and Lustrek, M. and

Gams, M. (2015). Automatic detection of perceived

stress in campus students using smartphones. Intelli-

gence Environment. 132-135

Hang, A. and Goronzy, S. and Schaich, P. and Williams,

J. (2014). Emotion Recognition Using Bio-Sensors:

First Steps Towards an Automatic System. Tutorial

and Research Workshop on Affective Dialogue Sys-

tems. 36-48

Kikamwa, R. and Liu, Y.X. and Lane, N.D. and Zhong, L.

(2013). MoodScope: Building a Mood Sensor from

Smartphone Usage Patterns. In Proc. of MobiSys13.1-

Koskimaki, H. et al (2017). Early detection of migraine at-

tacks based on wearable sensors: experiences of data

collection using Empatica E4. In Proc. of the 2017

ACM Int. Joint Conf. on Pervasive and Ubiquitous

Computing and the 2017 ACM Int. Symposium on We-

arable Computers (UbiComp ’17). 506-511

Laleh, J. and Dao, M.S. and Jain, R. and Zettsu, K. (2015)

Complex asthma risk factor recognition from hetero-

geneous data streams. In Proc. of IEEE Int. Conf. on

Multimedia and Expo Workshops (ICMEW)

Lewine, R. and Sommers, A. and Waford, R. and Robert-

son, C. (2015). Setting the Mood for Critical Thinking

INDEED 2018 - Special Session on INsights DiscovEry from LifElog Data

690

in the Classroom. Int. Journal for the Scholarship of

Teaching and Learning. 9 (2), Article 5, 1-4

Liew, T.W. and Tan, S.M. (2016). The Effects of Posi-

tive and Negative Mood on Cognition and Motivation

in Multimedia Learning Environment. Educational

Technology and Society. 19 (2), 104115

Mainhard, T. and Oudmana, S. an Hornstra, L. and Bosker,

R.J. and Goetz, T. (2017) Student emotions in class:

The relative importance of teachers and their inter-

personal relations with students. Learning and In-

struction. ISSN 0959-4752

Pekrun, R. (1992). The Impact of Emotions on Lear-

ning and Achievement: Towards a Theory of Cogni-

tive/Motivational Mediators. Applied Psychology: An

International Review. 41 (4), 359-376

Rachuri, K.K. et al. (2010). EmotionSense: A Mobile Pho-

nes based Adaptive Platform for Experimental Social

Psychology Research. In Proc. of UbiComp10. 281-

290

Robbins, S.P. and Judge, T.A. and Campbell, T. (2010)

Organizational Behaviour. Pearson. ISBN-10:

0273719394, ISBN-13: 9780273719397

Sano, A. and Picard, R.W. (2013). Stress Recognition using

Wearable Sensors and Mobile Phones. In Proc. of

Humaine Association Conference on Affective Com-

puting and Intelligent Interaction. 671-676

Sottilare, R.A. and Proctor, M. (2012). Passively Classi-

fying Student Mood and Performance within Intelli-

gent Tutors. Educational Technology and Society. 15

(2)

Thapliyal, H and Khalus, V. and Labrado, C. (2017). Stress

Detection and Management: A Survey of Wearable

Smart Health Devices. IEEE Consumer Electronics

Magazine. 6 (4). 64-69

Titsworth, S. and Quinlan, M.M. and Mazer, J.P. (2010)

Emotion in Teaching and Learning: Development and

Validation of the Classroom Emotions Scale. Commu-

nication Education. 59 (4), 431-452

Valenza, G. et al. (2014). Wearable Monitoring for Mood

Recognition in Bipolar Disorder based on History-

Dependent Long-Term Heart Rate Variability Analy-

sis. IEEE Journal of Biomedical and Health Informa-

tics. 18 (5), 1625-1635

Wioleta, S. (2013). Using Physiological Signals for Emo-

tion Recognition. In Proc. of 6th Int. Conf. on Human

System Interactions (HSI). 556-561

Zenonos, A. et al. (2016). HealthyOfﬁce: Mood Recogntion

At Work Using Smartphones and Wearable Sensors.

In Proc. of PERCOM Workshop. 1-6

HealthyClassroom - A Proof-of-Concept Study for Discovering Students’ Daily Moods and Classroom Emotions to Enhance a

Learning-teaching Process using Heterogeneous Sensors

691