Definitions/sensors-logo-eps-converted-to.pdf
Article
Towards the Applicability of Measuring the
Electrodermal Activity in the Context of Process
Model Comprehension: Feasibility Study
Michael Winter1* , Rüdiger Pryss2, Thomas Probst3, and Manfred Reichert1
1Institute of Databases and Information Systems, Ulm University, Ulm, Germany;
{michael.winter,manfred.reichert}@uni-ulm.de
2Institute of Clinical Epidemiology and Biometry, University of Würzburg, Würzburg, Germany;
ruediger.pryss@uni-wuerzburg.de
3Department for Psychotherapy and Biopsychological Health, Danube University Krems, Krems, Austria;
*Correspondence: Michael Winter; [email protected]
Received: date; Accepted: date; Published: date
Abstract:
Process model comprehension is essential in order to understand the Five Ws (i.e., who,
what, where, when, and why) pertaining to the processes of organizations. However, research in this
context showed that a proper comprehension of process models often poses a challenge in practice.
For this reason, a vast body of research exists studying the factors having an influence on process
model comprehension. In order to point research towards a neuro-centric perspective in this context,
the paper at hand evaluates the appropriateness of measuring the electrodermal activity (EDA)
during the comprehension of process models. Therefore, a preliminary test run and a feasibility study
were conducted relying on an EDA and physical activity sensor to record the EDA during process
model comprehension. The insights obtained from the feasibility study demonstrated that process
model comprehension leads to an increased activity in the EDA. Furthermore, EDA-related results
indicated significantly that participants were confronted with a higher cognitive load during the
comprehension of complex process models. In addition, the experiences and limitations we have
learned in measuring the EDA during the comprehension of process models are discussed in this
paper. In conclusion, the feasibility study demonstrated that the measurement of the EDA could be
an appropriate method to obtain new insights in process model comprehension.
Keywords: Process Model; Process Model Comprehension; Electrodermal Activity; Sensor
1. Introduction
Process models are abstractions (i.e., in terms of documentation, definition, and execution) from
the physical world representing objects, procedures, or issues [
1
]. Process models are particularly
used to conceptualize, determine, and describe procedures, technical systems, or the processes
of organizations [
2
]. Regarding the latter, process models constitute a composition of activities,
decisions, data, and resources from associated organizational processes in order to achieve a particular
objective (e.g., product or service) [
3
]. However, process models not solely document the processes
of organizations, they additionally offer opportunities to extract specific process information (e.g.,
key performance indicators) for the purpose of process analysis, optimization, and automation [
4
]. In
addition, process models provide means for collaboration purposes, thus facilitating conveyance of
process information between stakeholders [5].
In order to benefit from the application of process models, organizations must take care that the
proper understanding of such models (i.e., process model comprehension) is ensured for all involved
Sensors 2020,xx, 5; doi:10.3390/sxx010005 www.mdpi.com/journal/sensors
Sensors 2020,xx, 5 2 of 22
stakeholders [
6
]. Therefore, the identification of factors is vital, both positive and negative ones, which
are influencing the comprehension of process models. For example, if factors that hinder process
model comprehension are not addressed properly, respective processes might not deliver the required
results. Failures that happen in the application of such models have been commonly linked to model
incomprehension [7].
In this context, a large body of research has evolved over the last decade on studying the factors that
influence the comprehension of process models [
8
]. Therefore, different objective properties of process
models, such as the syntax [
9
], structure [
10
], labeling [
11
], coloring [
12
], visual notational deficiencies
[
13
], and secondary notation [
14
] were investigated, their influence on process model comprehension
elaborated, and corrective actions presented (e.g., guidelines [15]).
However, existing research in the context of process model comprehension addressed only
objective factors. Since different kinds of stakeholders (e.g., modeling and domain experts) are
involved in working with process models, their expertise in the comprehension of such models varies
[
16
]. Interestingly, research showed that expertise in working with process models is not the only
decisive factor influencing process model comprehension [
17
]. Moreover, despite existing research
in the context of process model comprehension, stakeholders, both experienced and inexperienced,
are still facing challenges on how to properly read and comprehend process models. Therefore, the
influence of subjective factors and their influence on model comprehension is addressed in recent
works. Examples are the modeling expertise [
18
], model reader preferences [
19
], learning strategies
[20], perceptual discrimination [21], or perceived usefulness [22].
As known from other domains, a focus is increasingly put on the influence of cognitive aspects [
23
].
For example, in the context of process model comprehension, the mental load as well as corresponding
efforts [
24
,
25
], cognitive style [
26
], cognitive biases [
27
], and cognitive load [
28
] are investigated.
Moreover, additional technologies are applied to get a deeper understanding about cognitive aspects
and their influence on the comprehension of process models. Prominently, the application of eye
tracking (e.g., [
29
–
32
]) is pursued. Additionally, the use of other technologies and methods (e.g., smart
mobile devices [33], serious games [34]) are becoming more popular in this context.
As such technologies (e.g., eye tracking [
35
]) become increasingly affordable and are fanned by
the proliferation of additional sensors (e.g., in smart mobile devices [
36
]), the identification of factors
influencing process model comprehension can thus be facilitated for research in a novel manner. More
specifically, the application of smart sensors allows for the analysis of factors not previously considered
in the context of process model comprehension. Examples could be the measurement of physiological
parameters (e.g., heart rate [
37
]) or psychological factors such as the level of arousal [
38
]. Regarding
the latter, research showed that the level of arousal and the different states (e.g., tension, excitement)
thereof affect our behavior (e.g., decision making [
39
]). Usually, the level of arousal (i.e., the state of
being awake and attentive) is measured either by self-reporting tools (e.g., questionnaires [
40
]) or
through the application of technologies (e.g., wearable sensors [
41
]) that are measuring body reactions
against environmental influences. Other examples are the consideration of pupil dilation with eye
tracking [
42
] or the brain activity with the Electroencephalography (EEG) [
43
]. Another approach for
the analysis of the level of arousal is the measurement of the skin conductivity, also known as the
electrodermal activity (EDA) [44–46].
For this reason, to the best of our knowledge, no works exist that considers the electrodermal
activity (EDA) in the context of process model comprehension. Therefore, this paper presents first
insights, experiences, and lessons learned gathered in EDA research (i.e., preliminary test run and
feasibility study) during the comprehension of process models. The emphasis was put on the research
question, which evaluated the applicability as well as the appropriateness of measuring the EDA
relying on a smart EDA and physical activity sensor during the comprehension of process models.
Moreover, the work at hand shall, on the one hand, foster further EDA studies in this context and, on
the other hand, shall contribute towards research to facilitate an in-depth neuro-centric perspective in
terms of process model comprehension.
Sensors 2020,xx, 5 3 of 22
The remainder of this paper is structured as follows: Section 2 introduces theoretical background
about the EDA. In addition, Section 2 describes the study context and the setting of the conducted
EDA research (i.e., test run and feasibility study). The obtained EDA results are presented, tested for
significance, and discussed in Section 3. Moreover, this sections presents revealed limitations and the
lessons we learned. Finally, Section 4 summarizes the paper and gives an outlook on future work.
2. Materials and Methods
2.1. Electrodermal Activity
The electrodermal activity (EDA) describes variations in the eccrine sweat gland production of
the human body. These variations in the sweat production result in changes of the electrical skin
properties (i.e., skin conductance) [
47
]. Notably, sweating is controlled by the sympathetic nervous
system (i.e., part of the autonomic nervous system) and changes in the skin conductance are indications
for physiological or psychological arousal (e.g., fight-or-flight response) [
48
]. Research demonstrated
that this kind of arousal is significantly related to brain functions that regulate motor, sensory, and
cognitive skills [
44
,
49
,
50
]. For example, when emotionally agitated (e.g., on the eve of an exam), the
sweat production is increased, resulting in an increase in the EDA as well (e.g., higher cognitive load).
In turn, at rest, sweat production and associated EDA is low. In general, the EDA is measured with the
application of sensors that are attached either on the sole feet or the palms. The reason for this way of
attachment is that the number of eccrine sweat glands is highest in these two places. Furthermore, the
EDA describes a raw electric signal that consists of two components characterizing the phasic skin
conductance response (SCR) and the tonic skin conductance level (SCL) [
51
]. More specifically, the
SCR constitutes abrupt increases in the skin conductance as a direct reaction towards an environmental
stimulus. Usually, these abrupt increases emerge between one to five seconds after the presentation
of a stimulus (e.g., picture, sound). Such increases are strongly associated with cognitive processes
(e.g., decision making) following a short-term event [
52
]. The SCR (specified in microsiemens,
µ
s) is
characterized by five factors over time after the appearance of a stimulus:
1
latency,
2
rise time,
3
amplitude,
4
peak, and
5
recovery time. While the latency has a usual duration between one and five
seconds, the duration of the other four factors is dependent on the individual as well as the presented
stimulus. An example of the SCR is depicted in Fig. 1. In addition, SCR exist that occur spontaneously
in the absence of any stimulus (i.e., Non-SCR) [47].
In contrast to phasic SCR, the tonic SCL (specified in microsiemens,
µ
s) is defined as the slowly
changing raw level of skin conductance. Changes in the SCL are not triggered by particular stimuli
or events, but represent a continuous intra-individual course over the period of time. The SCL
varies significantly between individuals and is affected by psychological states, physical condition,
Time (s)
EDA (µs)
Stimulus
Latency
Stimulus
(1) Latency
(4) Peak
(3) Amplitude
(5) Recovery Time
(2) Rise Time
Time (s)
EDA (µs)
Figure 1. Skin Conductance Response (SCR) after a Stimulus
Sensors 2020,xx, 5 4 of 22
and autonomic regulation. Moreover, the size of the used electrodes for measuring the EDA signal
is an additional influence factor. Although the phasic SCR is more prominent in EDA research,
insights revealed the importance of considering both components (i.e., SCR and SCL) in order to
better understand the physiological as well as psychological processes and their reactions towards
specific stimuli [
53
]. In this context, Fig. 2exemplarily outlines the distinction between SCR (i.e., phasic
component) and SCL (i.e., tonic component) in a raw EDA signal while writing an exam. The exam
situation leads to a slow and continuous increase in the SCL (i.e., blue), since the exam represents a
tense situation requiring an attentive state. While solving single tasks in the exam, there are repeatedly
abrupt increases in the SCR (i.e., green) over the time. These abrupt increases are indications for fast
and short-term amplifications of cognitive processes (e.g., reasoning, decision making). When the
exam is passed, and the caused tension declines, the SCL also decreases steadily towards a baseline
level.
Time (s)
EDA (µs)
SCR (Phasic)
SCL (Tonic)
Begin Exam End Exam
Figure 2.
Distinction of the Tonic Skin Conductance Level (SCL) and Phasic Skin Conductance Response
(SCR) in a Raw EDA Signal
2.2. Context Selection
In general, comprehension is a cognitive process that is strongly affected by the level of arousal
having an impact on, for example, reading [
54
], learning [
55
], and information processing [
56
].
Similarly, the comprehension of process models is a complex matter. On the one hand, there must
be an adequate level of knowledge about the process modeling notation used for the creation of
respective process models. On the other hand, documented information in process models need to be
decoded and captured properly by all stakeholders [
17
]. Existing research have already made major
contributions in the context of process model comprehension. However, the measurement of the EDA
has not been addressed so far in prior works. To address this gap, the paper at hand investigates the
following research question:
Research Question
Is the measurement of the EDA during process model comprehension an appropriate method in order to
foster our understanding of working with process models?
This paper presents the first insights towards measuring the EDA in the context of process model
comprehension. The insights obtained shall contribute to a novel neuro-centric perspective for research
on the comprehension of process models. Our existing conceptual framework for the comprehension
of process models that already incorporates methods and theories from cognitive neuroscience and
psychology is therefore enriched by the findings from this work [
57
]. A preliminary test run and a
feasibility study were conducted relying on the measurement of the EDA with a specific EDA and
physical activity sensor. The purpose of the test run was to familiarize ourselves with the measurement
of the EDA. This included the application of the measuring sensor device, execution of the EDA
Sensors 2020,xx, 5 5 of 22
measurement as well as the analysis and interpretation of the results. With the knowledge gained from
the test run, a feasibility study was conducted, in which participants needed to comprehend a set of
differently complex process models, while, at the same time, their EDA was recorded.
(a) Top Side of the EdaMove 3 Sensor (b) Electrodes Attached to the Palm
Figure 3. Attachment of the Sensor EdaMove 3
2.3. Instrumentation
For both phases (i.e., test run and feasibility study), the EDA and physical activity sensor EdaMove
3 was used [
58
]. The EdaMove 3 is a smart sensor device to record and analyze skin conductivity (i.e.,
EDA) as well as physical activity (e.g., step counter). The sensor has already been successfully applied
in different fields of research (e.g., psycho physiologic monitoring [
59
], affective computing [
60
]) and
provides all relevant standards for the proper measurement of the EDA and corresponding components
(i.e., phasic SCR, tonic SCL). In more detail, an exosomatic measurement with direct current (DC) (i.e.,
.5 V constant voltage system) is used for the assessment of the EDA signal. The measurement range is
between 0
µ
s and 100
µ
s, accuracy is < 1.5 %, and with a resolution of 14 bit. Additionally, the sensor
enables a live analysis of the EDA, body temperature, step count, and movement acceleration in all
dimensions. The EdaMove 3 has two reusable electrodes (i.e., non-polarizing sintered Ag/AgCl) with
an electrode disc diameter of 10 mm (i.e., 78,5 mm
2
effective electrode area). The electrodes can be
attached either on the sole feet or the palms with adhesive tape rings supported by an electrode gel
and alcohol in order to measure the EDA with a sample rate of 32 Hz (bandwith: DC to 8 Hz). In
the presented research, the sensor was attached to the palms (see Figs. 3(a) and (b)). The EdaMove
3 was configured with all relevant information (e.g., time for the start of the measurement) with the
Movisens SensorManager 1.14.4. Recorded sensor data was visualized and preprocessed for first
analyses using the UnisensViewer 1.12.38. Furthermore, DataAnalyzer 1.11.12 was used for data
transformation and the calculation of relevant measures (e.g., SCL, SCR). Note that the detection
parameters of DataAnalyzer are based on the proposed approaches in [
61
] and [
53
]. More specifically,
the tonic component of the EDA signal (i.e., SCL) was low pass filtered (i.e., second order butterworth)
with a filter frequency of .1 Hz. Phasic SCRs were detected automatically (i.e., second order high pass
butterworth filter with .1 Hz) under consideration of the following criteria:
1
default minimal rise
time for SCR detection >.05
µ
s/sec,
2
default minimal amplitude for SCR detection >.1
µ
s,
3
default
maximal rise time for SCR detection <.9 sec. Importantly, there is an exception in SCR peak detection,
which occurs with overlapping (i.e., superimposed) peaks. More specifically, instead of a signal drop
after reaching the peak (i.e., recovery time), an ascent of the signal caused by another SCR peak is
detected. In this special case, the first emerging peak can be determined with an amplitude >.05
µ
s [
61
].
In general, the parameters, which constituted a potential limitation (see Section 3.3), were defined from
the vendor of EdaMove 3 as a reasonable compromise in order to allow an offline as well as online
live analysis in this context . Further, the stimuli (i.e., pictures in the test run, process models in the
Sensors 2020,xx, 5 6 of 22
feasibility study (see Sections 2.4 and 2.5) were presented on a 23” monitor (resolution of 1920x1080,
96 PPI). Finally, SPSS 26 was used for all statistical analyses.
(a) Positive - Giant Panda (b) Neutral - Table (c) Negative - Straitjacket
Figure 4. Pictures from the Geneva Affective Picture Database
2.4. Preliminary Test Run
The purpose of the small scale preliminary test run was to familiarize ourselves with the
measurement of the EDA. This included the accurate application of the EdaMove 3, correct
measurement of the EDA signal as well as respective components (i.e., SCL, SCR), and the correct
analyses of the recorded EDA data. Therefore, n = 4 participants had to contemplate various pictures
obtained from the Geneva Affective Picture Database (GAPED) [
62
]. Note that all participants gave
their informed consent for inclusion before they participated in the test run. GAPED contains a total of
730 pictures in terms of emotion induction. The pictures represent either humans, animals, or objects
that are related to positive, negative, or neutral emotions. These pictures were rated regarding arousal,
valence, and accordance with internal and external norms. Fig. 4exemplarily presents pictures that
should arouse positive (i.e., (a) giant panda), neutral (i.e., (b) table), or negative (i.e., (c) straitjacket)
emotions. In order to improve our experience with the EdaMove 3 and the measurement of the EDA,
the pictures from GAPED were shown to the participants in short trials with different settings. For
example, the display duration of the pictures varied (from 1 up to 15 seconds), the periods of rest in
order to identify a proper baseline level to start with the measurement were between 1 and 15 minutes,
and participants needed to indicate whether a picture triggered rather positive, negative, or neutral
emotions. Fig. 5presents an excerpt of the EDA signal measured with EdaMove 3 (i.e., UnisensViewer).
The figure depicts the raw EDA signal (i.e., no distinction between tonic SCL and phasic SCR) showing
the baseline measurement (i.e., from minute 0 to 4) as well as the presentation of a positive (i.e., from
minute 4) and neutral (i.e., from minute 4.30) picture from GAPED. Generally, the EDA findings that
we have gained from the test run were in accordance with existing literature (e.g., EDA was higher
in positive-related pictures compared to neutral ones) [
47
,
63
,
64
]. Furthermore, the experiences we
gathered and the lessons we learned from the test run are included in the discussion presented in
Section 3.4.
Figure 5. Excerpt from an EDA Measurement
Sensors 2020,xx, 5 7 of 22
2.5. Feasibility Study
For the investigation of the research question (see Section 2.2), using the gained experiences from
the test run, a feasibility study was conducted, which was performed as follows:
Participants:
The feasibility study included a total of n = 9 participants. All participants were
students (i.e., 2 were female, all participants were under 25 years) and the study was conducted at
Ulm University. Prior to the feasibility study, all participants gave their informed consent for inclusion.
The participants were divided into two groups (i.e., Groups A and B) using the round-robin approach
(i.e., alternating assignment to Groups A and B). Group A consisted of n = 5 and Group B of n = 4
participants.
Text
Office
Define
Survey
Start
Survey Choose
Participants
Invite
Participant Interview
Participant Evaluate
Survey End
Survey
Schedule
Survey No
Yes
Inteview more
Participants?
Employee
(a) Easy Process Model - Implementation of a Survey
Sales
Order Department
ProductionShipping
Accept Order
Incoming
Order
Forward
Order
Check
Availability
Order
Materials
Produce
Goods
Document
Quality Flaw
Correct
Quality Flaw Review
Goods
Ship Order
Send Invoice
Complete
Order
No
Yes
Yes
No
Goods in
Stock?
Quality is
Fine?
(b) Complex Process Model - Order for Goods
Figure 6. Used Process Models in the Feasibility Study
Process Modeling Elements
Process Model Activity Event Gateway Edge Pool Lane Total
Easy 6 2 4 13 1 1 27
Complex 10 2 6 19 1 3 42
Table 1. Number of Modeling Elements in the Easy and Complex Process Model
Materials:
Two differently complex process models (i.e., easy and complex) expressed in terms
of the Business Process Model and Notation (BPMN) 2.0 were used in this study [
65
]. The process
models were composed of basic elements of BPMN 2.0. Note that the easiest model showed the
implementation of a survey. In turn, an order for goods was presented in the complex one. Figs. 6
show both process models: Fig. 6(a) presents the survey scenario, while Fig. 6(b) the order for goods
1
.
More specifically, in order to ensure differences in process model complexity, the two models were
created based on the metrics proposed in [
66
]. Therefore, the two process models varied in the number
1High-resolution images of the used process models are available here: https://tinyurl.com/y7g5hgy2
Sensors 2020,xx, 5 8 of 22
of modeling elements and outgoing sequence flows. The following Table 1presents the number of
process modeling elements, each used in the two respective models.
Measures:
Although various parameters (e.g., SCR energy, recovery time) can be considered in
EDA analyses [
47
], for a first evaluation of the applicability of the EDA in the context of process model
comprehension, the following three EDA-related measures were considered in the feasibility study:
•Mean SCL:
The tonic SCL describes the changing level of skin conductivity over the period of
time (see Section 2.1). A state of physiological or psychological arousal usually leads to variations
(e.g., increase) in the SCL. In our context, the comprehension of process models constituted a
cognitively challenging task that created a state of arousal (e.g., attentive). Therefore, the analysis
of the SCL allowed for the assumption whether the comprehension of differently complex process
models results in variations (e.g., elevation) in the SCL.
•Number of SCR Peaks:
SCR peaks are parts of the phasic component that are indications for
short-term processes with high physiological (e.g., the wait for the go-ahead) or psychological
(e.g., decision making) demands. In process model comprehension, the correct interpretation
of process information must be ensured and, therefore, decisions (e.g., which activities may
run in parallel) must be made, which are decisive for the perception as well as the correct
comprehension of the process model. For this reason, it was interesting to evaluate whether the
number of SCR peaks was higher in the complex process model juxtaposed to the easy model.
Importantly, only SCR peaks with an amplitude height of > .1
µ
s were considered (i.e., special
case with > .05 µs; see Section 2.3).
•Mean of SCR Amplitudes Energy Level:
The mean of SCR amplitudes energy level is a measure
in order to record the degree of stress (e.g., cognitive load) a stimulus or event provokes. The
higher perceived stress is, the higher is the amplitude and vice versa. In our study, the evaluation
of the SCR amplitudes revealed insights about the cognitive load and related processes during
the comprehension of process models.
Study Design:
The design of the feasibility study was based on the guidelines proposed in [
67
].
The study was conducted in a prepared lab at Ulm University. The lab was quiet, ambiently dimmed,
and care has been taken to keep the lab temperature around 22 degree Celsius. Such preparations were
necessary in order to ensure the same study setting, since environmental influences (e.g., temperature)
have an impact on the EDA. The experience we gained from the preliminary test run (see Section
2.4) have contributed to the study design. Due to the availability of only one sensor device, only
one participant could be evaluated and each study session took about 25 minutes. A study session
was as follows: The participant was welcomed and the study procedure was explained. Afterwards,
informed consent as well as demographic information were provided. Following this, participants
were assigned into Group A or Group B using the round-robin approach (i.e., alternating assignment
to Group A or B) in order to ensure a balanced distribution in both groups across all participants.
Group A started with the comprehension of the easy process model and then the complex one. In turn,
Group B first had to comprehend the complex model, followed by the easy one. Then, the EdaMove 3
was attached to the palm (see Section 2.2). After the attachment of the sensor device, the participant
was asked not to talk for the remaining duration of the study. After completing all these steps, a first
baseline measurement was made. The participant was advised to sit comfortably, remain calm, and
relax for a total of ten minutes. Research has shown that in a state of relaxation, the EDA drops and a
baseline, from which the EDA measurements can be started, is reached [
47
,
51
]. Such arrangements
(e.g., not to talk, relaxation) are necessary in order to avoid potential external influences, which may
have an impact on the EDA signal. After ten minutes, either the complex (i.e., Group B) or the easy (i.e.,
Group A) process model was shown to the participant for a total of 30 seconds. In these 30 seconds,
the participant should comprehend the presented process model syntactically as well as semantically.
Subsequently, after comprehending the first model, a second baseline measurement was made in order
to ensure that the EDA drops towards the baseline level again. After the second baseline measurement,
Sensors 2020,xx, 5 9 of 22
the second process model (i.e., either the complex (i.e., Group A) or easy (i.e., Group B)) one was
shown for comprehension for another 30 seconds. In total, one EDA measurement was obtained from
each participant that was divided in four parts:
1
baseline measurement 1,
2
baseline measurement
2,
3
comprehension of the easy process model, and
4
comprehension of the complex process model
(see Section 3). Finally, Fig. 7summarizes the study design.
Introduction EDAMove 3
Attachement
Complex
Process
Model
Easy
Process
Model
Easy
Process
Model
2. Baseline
Measurement
1. Baseline
Measurement
1. Baseline
Measurement
Complex
Process
Model
2. Baseline
Measurement
Round-robin
Describe the procedure
of the study Attachement
to the palms
No talk at
this point
t = 10 Min. t = 30 Sec. t = 10 Min. t = 30 Sec.
t = 10 Min. t = 30 Sec. t = 10 Min. t = 30 Sec.
Group A (n = 5)
Group B (n = 4)
Reach state of relaxation Comprehend
process model
Figure 7. Study Design Used in the Feasibility Study
3. Results
Fig. 8shows the recording of a raw EDA signal measurement (i.e., without a separation of the
tonic and phasic EDA components SCL and SCR) from a participant of Group A. In more detail, the
figure depicts the two baseline measurements as well as the presentation of the easy and the complex
process model (PM). It can be seen from the figure how the EDA signal reaches a baseline within
the first ten minutes and makes a clear burst increase after presenting the first process model for
the purpose of comprehension to the participant. Afterwards, in the second baseline measurement,
the EDA signal shows up to be a little bit more unsettled juxtaposed to the EDA signal in the first
baseline measurement. A reason for this behavior of the EDA signal might be the level of arousal,
which is the physiological and psychological state responsible for different behavioral and cognitive
processes, such as attention, decision making, and information processing. It can be assumed that after
the comprehension of the first process model, the participant may have been busy recapitulating and
processing the process information just presented in the first model.
Figure 8. Presentation of a Raw EDA Signal
From Fig. 8, the tonic part of the EDA signal (i.e., SCL) is clearly visible. However, the distinction
of the phasic part (i.e., SCR peaks) is hardly possible from this observation. For this reason, the two EDA
components (i.e., SCL and SCR peaks) were considered separately by applying data transformation
(i.e., signal decomposition) provided by the application DataAnalyzer (see Section 2.3). Note that the
visualization of the phasic SCR component is currently limited in DataAnalyzer (see Section 3.3). Fig. 9
presents exemplarily the considered measures from the feasibility study (see Section 2.5) after the data
transformation. The raw
1
EDA signal is split into the
2
tonic SCL, and
3
phasic SCR amplitudes
energy level. Regarding the latter, corresponding SCR peaks (i.e., 8 in total) and related energy level (in
µ
s) are well recognizable and visualized as abrupt increases in this figure. Note the small periodic shift
Sensors 2020,xx, 5 10 of 22
Figure 9.
Analysis of the EDA Measures:
1
EDA Signal,
2
SCL, and
3
SCR Amplitudes Energy Level
of about 1 second in
2
SCL that was caused due to filtering. In more detail, the SCL was extracted
from the raw EDA signal by means of a second order butterworth filter (i.e., the output signal is shifted
in time with respect to the input signal). In order to address a signal delay, further signal processing
methods (e.g., IIR filter) should be applied (see Section 3.3).
The following tables present the results obtained from the feasibility study for the easy (see Table
2) and the complex (see Table 3) process model. For each participant (P) in the respective group (i.e.,
Group A or B), mean and standard deviation for the three considered measures (M, see Section 2.5)
Mean SCL (in
µ
s), Number of SCR Peaks, and Mean of SCR Amplitudes Energy Level (in
µ
s) are
shown in the tables. Further, only SCR amplitudes exceeding the threshold of >.1
µ
s were considered
(i.e., special case with >.05
µ
s; see Section 2.3). Moreover, the averages for all measures obtained from
all participants for the respective group are shown in both tables.
Table 2. Descriptive Results for the Easy Process Model from the Feasibility Study
Group A Group B
HHHH
H
P
MMean SCL SCR
Peak
SCR
Amp Mean SCL SCR
Peak
SCR
Amp
1 11.12 (.49) 5.00 .21 (.09) 5.83 (.70) 6.00 .75 (.27)
2 4.83 (.32) 7.00 .58 (.32) 2.82 (.13) 4.00 .15 (.09)
3 3.16 (.15) 3.00 .18 (.12) 4.82 (.14) 5.00 .23 (.10)
4 2.79 (.20) 6.00 .29 (.08) 4.87 (.26) 4.00 .58 (.31)
5 2.01 (.13) 4.00 .12 (.06) – – –
Avg 4.78 (3.31) 5.00 (1.41) .28 (.25) 4.59 (1.16) 4.75 (.83) .43 (.33)
Note: P = Participant; M = Measure; SCL = Skin Conductance Level; SCR = Skin
Conductance Response; Amp = Mean Amplitudes Energy Level
Table 3. Descriptive Results for the Complex Process Model from the Feasibility Study
Group A Group B
HHHH
H
P
MMean SCL SCR
Peak
SCR
Amp Mean SCL SCR
Peak
SCR
Amp
1 10.44 (.1) 4.00 .35 (.06) 6.46 (.24) 9.00 .30 (.15)
2 4.45 (.28) 6.00 .56 (.23) 2.74 (.19) 7.00 .26 (.23)
3 2.24 (.05) 8.00 .12 (.04) 4.90 (.43) 10.00 .34 (.21)
4 3.04 (.01) 6.00 .24 (.12) 2.98 (.18) 5.00 .29 (.09)
5 1.83 (.11) 8.00 .19 (.10) – – –
All 4.40 (3.15) 6.40 (1.50) .29 (.20) 4.27 (1.54) 7.75 (1.92) .29 (.19)
Note: P = Participant; M = Measure; SCL = Skin Conductance Level; SCR = Skin
Conductance Response; Amp = Mean Amplitudes Energy Level
Moreover, the Tables 4,5, and 6present mean and standard deviation (i.e., M (SD)) of the three
obtained measures (see Section 2.5) during the two baseline measurements. More specifically, in
Table 4, for each process model (i.e, easy and complex), the Mean SCL (in
µ
s) obtained in respective
Sensors 2020,xx, 5 11 of 22
baseline measurement (BM) from each participant (P) of Groups A and B before the comprehension of
respective model is shown as well as the aggregated results of all participants from respective group.
In Table 5, for each 30 second time interval during the baseline measurements, the Number of Non-SCR
Peaks (i.e., absence of stimulus) from all participants and the aggregated results are shown. Finally,
Table 6depicts Mean of SCR Amplitudes Energy Level (in
µ
s, with threshold >.1
µ
s, special case with
>.05 µs) for each participant and the average of all participants.
Table 4. Descriptive Results for the SCL during Baseline Measurements
Easy Process Model Complex Process Model
P BM SCL P BM SCL P BM SCL P BM SCL
A1 11.64 (.45) B1 5.42 (.78) A1 10.60 (.40) B1 5.14 (.52)
A2 4.12 (.43) B2 2.87 (.22) A2 4.10 (.47) B2 2.17 (.39)
A3 3.53 (.52) B3 4.66 (.69) A3 2.51 (.23) B3 3.72 (.84)
A4 2.47 (.28) B4 4.19 (.67) A4 2.64 (.36) B4 2.88 (.61)
A5 2.14 (.27) - - A5 2.18 (.53) - -
All 4.56 (2.74) All 3.99 (2.56)
Note: P = Participant; BM = Baseline Measurement; SCL = Skin
Conductance Level
Table 5. Descriptive Results for the Number of Non-SCR Peaks during Baseline Measurements
Easy Process Model Complex Process Model
P BM SCR P BM SCR P BM SCR P BM SCR
A1 5.05 (.74) B1 5.75 (.94) A1 3.80 (.75) B1 4.55 (1.40)
A2 4.15 (1.01) B2 4.20 (.79) A2 4.15 (.96) B2 4.65 (1.01)
A3 2.55 (.74) B3 4.75 (.83) A3 5.05 (1.24) B3 4.50 (1.86)
A4 5.30 (1.68) B4 3.65 (.79) A4 4.60 (1.59) B4 4.70 (1.42)
A5 3.95 (1,16) - - A5 4.9 (1.3) - -
All 4.39 (.95) All 4.55 (1.30)
Note: P = Participant; BM = Baseline Measurement; SCR = Number of
Non-Skin Conductance Response Peaks
Table 6. Descriptive Results for the SCR Amplitudes Energy Level during Baseline Measurements
Easy Process Model Complex Process Model
P BM Amp P BM Amp P BM Amp P BM Amp
A1 .22 (.09) B1 .47 (.19) A1 .33 (.14) B1 .24 (.09)
A2 .55 (.18) B2 .20 (.05) A2 .60 (.18) B2 .26 (.15)
A3 .16 (.04) B3 .23 (.12) A3 .18 (.07) B3 .31 (.12)
A4 .26 (.08) B4 .53 (.17) A4 .24 (.08) B4 .29 (.12)
A5 .15 (.05) - - A5 .20 (.07) - -
All .31 (.11) All .27 (.11)
Note: P = Participant; BM = Baseline Measurement; AMP = Skin
Conductance Response Amplitudes Energy Level
In general, the results we obtained from the feasibility study were in line with results from EDA
research in different fields [
47
,
68
,
69
]. More specifically, during the baseline measurements and the
comprehension of the process models, inter- and intra-individual variations in the Mean SCL were
encountered. The box plots shown in Figs. 10 and 11 are demonstrating these SCL variations (in
µ
s).
In more detail, the box plots in Fig. 10 are showing the SCL obtained from the participants during
the baseline measurements (see Section 2.5) before the comprehension of the easy (see Fig. 10 (a)) and
complex (see Fig. 10 (b)) process model. In turn, Fig. 11 presents the SCL from the participants during
the comprehension of the easy (see Fig. 11 (a)) and complex (see Fig. 11 (b)) process model. Regarding
the inter-individual variations, the SCL in the participants reflected distinct differences in the skin
conductivity (see Section 2.1) during the baseline measurements as well as the comprehension of the
process models. In terms of inter-individual variations, participants showed differences (e.g., general
Sensors 2020,xx, 5 12 of 22
elevation in the SCL during the comprehension of the process models) in their respective SCL during
the measurement of the baseline as well as in model comprehension.
Participant B4B3B2B1A5A4A3A2A1
SCL (µs)
12
11
10
9
8
7
6
5
4
3
2
7
8
9
10 11
12
253
254
Page 1
(a) Easy Process Model
Participant B4B3B2B1A5A4A3A2A1
SCL (µs)
12
11
10
9
8
7
6
5
4
3
2
468
467 466465
464
472
463
Page 1
(b) Complex Process Model
Figure 10. SCL during Baseline Measurements before Process Model Comprehension
Participant B4B3B2B1A5A4A3A2A1
SCL (µs)
12
11
10
9
8
7
6
5
4
3
2
3
2
1
5
4
3
1
2
Page 1
(a) Easy Process Model
Participant B4B3B2B1A5A4A3A2A1
SCL (µs)
12
11
10
9
8
7
6
5
4
3
2
2
1
11
10 12
3
2
1
Page 1
(b) Complex Process Model
Figure 11. SCL during Process Model Comprehension
Considering the Number of SCR Peaks (i.e., phasic component), the results (see Tables 2and
3) showed that more SCR peaks appeared during the comprehension of the complex process model
in both groups juxtaposed to the easy model. Hence, the results indicated a higher cognitive load
in related cognitive processes (e.g., reasoning, decision making) during the comprehension of the
complex process model. Moreover, considering Groups A and B, their results regarding the SCR peaks
were similar for the easy process model. However, regarding the complex model, the individual results
varied. The number of SCR peaks was higher in Group B compared to Group A. An explanation
might be that participants from Group A, who have seen the easy process model firstly, were already
prepared to comprehend the second complex model. In turn, Group B needed to comprehend the
complex process model firstly and, consequently, were subjected to more cognitive stress in order to
comprehend the first model properly.
In the context of the SCR Amplitudes Energy Level (see Tables 2and 3), except for Group B in
the easy process model, the amplitudes were at an average energy level of about .30
µ
s. Research
demonstrated that the average amplitudes energy level varies between .20
µ
s and .60
µ
s, which are
references for various cognitive processes (e.g., reasoning) [70].
In the context of the two baseline measurements (see Tables 4,5, and 6), the obtained results
assumed similar values for the three considered measures. Moreover, similarities in the Mean SCL
and Mean SCR Amplitudes Energy Level with the results obtained during the comprehension of the
process models (see Tables 2and 3) were discernible. Regarding the Number of SCR Peaks in the
easy process model, the average number was slightly higher during process model comprehension
compared to the baseline measurements. Furthermore, in the complex process model, the Number of
Sensors 2020,xx, 5 13 of 22
SCR Peaks differed notably (i.e., higher during model comprehension) juxtaposed to the number in the
baseline measurements. As a result, it appears that the Number of SCR Peaks was the only measure
indicating distinctions (i.e., inter-individual) during the comprehension of differently complex process
models.
Finally, Table 7presents the aggregated results (i.e., mean and standard deviation) for the easy and
complex process model regarding the three considered EDA measures (i.e., Mean SCL (in
µ
s), Number
of SCR Peaks, Mean SCR Amplitudes Energy Level (in
µ
s); see Section 2.5) from the participants of
Groups A and B.
Table 7. Descriptive Results for the Easy and Complex Process Model
PM PM SCL SCR Peak SCR Amp
Easy 4.70 (2.56) 4.89 (1.20) .37 (.30)
Complex 4.34 (2.57) 7.00 (1.83) .29 (.19)
Note: PM = Process Model; SCL = Skin Con-
ductance Level; SCR = Skin Conductance Re-
sponse; Amp = Amplitudes Energy Level
Regarding the Mean SCL during process model comprehension and the SCR Amplitudes Energy
Level, only small differences can be seen (see Table 7). However, it appears that the Number of SCR
Peaks was higher during the comprehension of the complex process model juxtaposed to the easy one.
As with previous observations (see Tables 2–6), the results for the three measures in Table 7reflected
similar average values, but the Number of SCR Peaks in the complex process model further confirmed
the indication as a measure in order to observe distinctions during process model comprehension of
varying model complexity.
3.1. Inferential Statistics
To evaluate whether the differences between the easy and complex process models seen in the
descriptive results (see Table 7) reach statistical significance, the Wilcoxon signed-rank test for two
related samples was performed for the three considered EDA measures (i.e., Mean SCL, Number of
SCR Peaks, Mean of SCR Amplitudes Energy Level; see Section 2.5). All statistical tests were performed
two-tailed and the significance value was set to p < .05. Additionally, for each measure in respective
process model, mean (M), standard deviation, and median (Mdn) is reported.
•Mean SCL:
The Wilcoxon signed-rank test indicated that the mean SCL in the complex process
model (M = 4.34 (2.71), Mdn = 3.04) was not significantly higher than the mean SCL in the easy
process model (M = 4.34 (2.72), Mdn = 4.82), Z=−1.362, p<.173.
•Number of SCR Peaks:
The Wilcoxon signed-rank test indicated that the number of SCR peaks in
the complex process model (M = 7.00 (1.94, Mdn = 7.00)) was significantly higher than the number
of SCR peaks in the easy process model (M = 4.89 (1.27), Mdn = 5.00), Z=−1.975, p<.048.
•Mean of SCR Amplitudes Energy Level:
The Wilcoxon signed-rank test indicated that the mean
of SCR amplitudes energy level in the complex process model (M = .29 (.12), Mdn = .29) was not
significantly higher than the mean of SCR amplitudes energy level in the easy process model (M
= .34 (.23), Mdn = .23), Z=−.059, p<.953.
In conclusion, the differences in the Mean SCL between Groups A and B was not significant.
Reasons could have been, on the one hand, the large disparity in the SCL between participants (i.e.,
inter-individual), which is associated with the application of such measurement (see Section 2.1), and,
on the other, the equally distributed heterogeneity of intra-individual SCL variations in both groups.
However, the Number of SCR Peaks was significantly different and more SCR peaks occurred in
the comprehension of the complex process model, which implied an agitated level of arousal (e.g.,
higher cognitive load) compared to the easy model. Finally, the Mean of SCR Amplitudes Energy
Level showed no significant differences, which indicated that the participants were concerned with
Sensors 2020,xx, 5 14 of 22
the correct comprehension of respective process models. Summarizing, inferential statistics confirmed
that the Number of SCR Peaks was the only measure in the feasibility study to determine distinctions
during the comprehension of differently complex process model.
3.2. Discussion
The focus of this work was to answer the research question (see Section 2.2) whether the
measurement of the EDA relying on a smart sensor (i.e., EdaMove 3) during process model
comprehension is an appropriate method to foster our general understanding of working with such
models. The results obtained from the feasibility study (see Section 3) indicated that participants from
both groups (i.e., Groups A and B) were in a higher state of cognitive arousal during the comprehension
of process models. Moreover, the results regarding the tonic SCL revealed why in most research only
the phasic SCR is more prominent (see Figs. 10 and 11) [
47
]. Since changes in the SCL are due to
different factors and may vary inter- and intra-individual (see Section 2.1), their interpretation is often
difficult. For example, whether a mean SCL of 6 is low, average, or high depends on the individual
(similar for the SCR) [
53
]. In our context, similar to the work presented in [
71
], it could be shown
that the comprehension of process models stimulated the level of arousal, which consequently led to
noticeable variations (i.e., inter- and intra-individual) in the SCL from each participant over a period of
time. However, especially inter-individual variations were consistent when observed repeatedly over
a longer period of time. In the context of intra-individual variations, since all participants were not
engaged on any other activity than comprehending the process models, the variations obtained during
the comprehension of the easy and complex process model were alike. In general, during process
model comprehension, not only the semantic process information must be comprehended properly,
but it must be also ensured that the syntactic information of respective process modeling notation
(i.e., BPMN 2.0) is correctly comprehended. In this context, comprehension is a psychological process,
which can be measured with the consideration of the EDA [
72
]. More specifically, comprehension leads
to an increase in the EDA due to higher activation in the eccrine sweat gland production. Moreover,
the appearance of significantly more phasic SCR peaks (see Section 3.1) in the complex process model
revealed that participants were under higher cognitive load during model comprehension. The
reason is that in the complex process model (see Fig. 6(b)) more information must be interpreted
and comprehended correctly compared to the easy model (see Fig. 6(a)). Furthermore, this measure
(i.e., Number of SCR Peaks) was identified in the feasibility study as the only measure, which was
able to recognize significant distinctions during the comprehension of the differently complex process
model (i.e., easy and complex). The obtained SCR amplitudes energy levels indicated that participants
(i.e., inter- and intra-individual) were mainly occupied with the comprehension of the presented
process models and related cognitive process (e.g., reasoning, decision making) [
70
]. Higher SCR
amplitudes energy levels are particularly evident in aversive situations (e.g., fear) or during heavy
physical exertion (e.g., weightlifting). In addition, comparing the three measures during process model
comprehension with the ones obtained from the two baseline measurements, the latter showed minimal
differences (i.e., lower values). However, the results confirmed the indication that the Number of SCR
Peaks was the only significant measure in the feasibility study in order to measure process model
complexity. In general, our results in the context of process model comprehension were in line with
related EDA results obtained in other studies with different emphases [
47
,
53
,
73
–
75
]. Moreover, similar
to the correlation between the EDA and the difficulty in tasks shown in the work of [
69
], there might
be a correlation between EDA and process model comprehension resulting in a correlated elevation
of the EDA with increasingly complex process models (see Section 4). While in single short-term
observations within seconds, the interpretation of the emergence of a SCR peak can be attributed
to the cause (e.g., presentation of an external stimulus). However, in long-term observations, such
attribution only by measuring the EDA is difficult. Referring to Fig. 9, a total of 8 SCR peaks are
shown during the 30 second presentation of the process model. Reasons for the appearance of the SCR
peaks could have been manifold (e.g., classification or revision of comprehended process information),
Sensors 2020,xx, 5 15 of 22
but in the end, only assumptions about their appearance can be made. An approach allowing for a
better interpretation and to better support the causality would be the collection of more information
(e.g., self-reporting tools) and further parameters with additional technologies (e.g., smart sensors)
during the measurement of the EDA [
63
]. For example, the recording of eye movements may assist in
a better interpretation of the SCR peaks. With the correlation of the eye movements at the time of a
SCR peak, more concrete interpretations about the reasons for the emergence of a SCR peak can be
derived. Moreover, another challenge that needed to be considered during the EDA measurement is
demonstrated in Fig. 12. More specifically, the figure visualizes the raw EDA signal of a participant
from Group B. At first, similar to Fig. 8, the first baseline measurement leads to a stabilization of the
EDA signal towards a baseline level. However, from minute 6 on in the first measurement and from
minute 14 on in the second one, a significant and unsettled increase in the tonic (i.e., SCL) as well as in
the phasic (i.e., SCR) EDA signal were recognizable. Especially in the second baseline measurement,
the peaks were evident, although, as with the other participants in the feasibility study, the study
setting was the same. However, it can only be conjectured what the possible reasons were (e.g., external
influences). Consequently, it is important to keep an eye on such variations as they may affect the
actual EDA measurement.
Figure 12. Challenges in Analyzing the EDA Signal
Summarizing, we demonstrated successfully in a feasibility study that the measurement of the
EDA with the application of a specific EDA and physical activity sensor (i.e., EdaMove 3) during
process model comprehension is an appropriate method in order to foster research in this context.
Although there were several limitations and aspects that needed to be considered carefully (see Sections
3.3 and 3.4), the measurement of the two EDA components (i.e., tonic SCL and phasic SCR) provides
interesting insights to support our understanding of working with such models. Research as well as
practice may benefit both in the future from the obtained insights. For example, a better support in
model comprehension with the definition of rules as well as directives ensuring a proper process model
comprehension can be pursued. Moreover, to know how a stakeholder reacts in terms of cognitive
load during the comprehension of a process model allows for predictive analytics for a more focused
assistance adapted to individual needs. Furthermore, tool support can be individually improved with
knowledge about physiological or psychological factors (e.g., visualization) that influences process
model comprehension. Finally, additional research can focus on psychological aspects pertaining to, for
example, information processing, reasoning, and decision making in order to enhance our knowledge
in the comprehension of process models from a neuro-centric perspective.
3.3. Limitations
Although first results seem to be promising from the feasibility study, their generalization needs
to be confirmed either by replication or similar studies. In particular, several limiting factors were
encountered in the feasibility study that needed to be discussed. First, usually process models
document complex processes of the real world. In turn, the process models used in the feasibility
study were of simple nature. As a result, complex process models require more cognitive effort for a
proper comprehension, which may have led to differences in the EDA data compared to the EDA data
we obtained in the feasibility study. Second, the time (i.e., t = 30 sec.) set for the comprehension of
both process models constitute another limitation. For the easy process model, participants could have
Sensors 2020,xx, 5 16 of 22
completely comprehended the model before the time was over and, hence, the EDA signal afterwards
did not correspond to the comprehension of the process model. In turn, regarding the complex
process model, the time may have been set too short, which may have caused additional stress for the
participants. Third, another limitation were the participants of the feasibility study. On one hand, the
sample size was small (n = 9) and only students were evaluated, which limits generalizability. Fourth,
as we did not have a special lab, in which exact laboratory conditions can be always prevailed, the EDA
measurements may have been affected due to differences in the these environmental conditions. Fifth,
we did not ask the participants in the feasibility study about their physiological and psychological
condition. The EDA measurement is very sensitive and, hence, different conditions (e.g., tiredness due
to poor sleep) may have affected the EDA signal. Sixth, although the sample rate (i.e., 32 Hz) of the
sensor EdaMove 3 is adequate for most applications, however, fine-grained EDA signals may have
been not recorded during the measurements in the feasibility study. This is a result of a compromise
to enable an adequate offline as well as online live analysis. Seventh, in the context of EDA signal
decomposition, restricted visualizations (e.g., no complete phasic SCR component illustration), missing
signal corrections (e.g., no signal interpolation after filter application), and fixed parameters (e.g., >.1
µ
s as SCR amplitudes threshold; >.05
µ
s is recommended in literature [
45
]) also limited data analysis.
Moreover, there might be noise as well as further significant differences in the EDA data between the
comprehension of the easy and complex process model, which we could not show in the feasibility
study (e.g., loss of information), but which might become apparent with more accurate parameters of
higher resolution (e.g., SCL low pass filter with a lower cut-off frequency than .1 Hz).
3.4. Lessons Learned
In this section, the experiences we gathered in the measurement of the EDA from the conducted
research (i.e., preliminary test run and feasibility study) are summarized. In general, they constitute
valuable lessons learned that will allow for optimizations of similar future studies in the context of
process model comprehension:
•
Baseline measurement: The baseline represents the average skin conductance level during rest
and without the presence of any stimulus. Moreover, the baseline varies over time depending
on various factors (i.e., physiological or psychological arousal). Therefore, it is of importance to
identify a baseline level for each individual separately before the start of an EDA measurement.
There are different recommendations regarding the duration of the baseline measurement, but
most of the research recommend a duration between 10 and 15 minutes [
47
,
51
]. In our studies,
we could observe that the EDA signal stabilized at a low level after about 8 minutes. In addition,
the baseline measurement can be used for a more fine-grained analysis of the EDA. For example,
individuals can be identified that are hyper- or hypo-responders to a stimulus. Further, during
relaxation, the identification of the frequency of Non-SCR (see Section 2.1) is simplified [76].
•
Recording of both EDA components: The initial research only considered the phasic SCR, while
the tonic SCL was not taken into account. For short-term observations (e.g., neural reaction),
the SCL can be neglected. In turn, for long-term observations, both EDA components should
be recorded, since both rely on different neural mechanisms. In our context, the consideration
of both components allowed for the interpretation that the comprehension of process models
resulted in a state of higher cognitive arousal. Finally, with the SCR, we were able to show that
the comprehension of a complex process models requires more cognitive effort.
•
Limit physical activity: The EDA is a very sensitive signal and even small movements (e.g.,
finger movement) may cause changes in the respective signal. Depending on the accuracy of the
used EDA sensor device, even contemplation may change the EDA signal. Therefore, in order to
avoid such changes, we ensured that the participants in our studies did not have to perform any
additional activities and could, therefore, concentrate on the comprehension of the presented
process models.
Sensors 2020,xx, 5 17 of 22
•
Avoid external stimuli: Similar to the activity limitation, any external stimuli (e.g., crowing bird,
light changes) may affect the EDA signal: several times we could observe this affection in the test
run as well (e.g., voices in the other room). Therefore, we have accepted this and tried to avoid
external stimuli. Hence, the recommendation is to conduct further EDA measurements in special
labs (e.g., light and soundproof) to ensure a proper recording of respective EDA components.
•
Constant setting: Another important factor that needs to be considered in the measurement of
the EDA is to keep a constant setting across all participants. In particularly, this ensures a valid
comparability of the recorded EDA signals obtained from all participants. In this context, among
others, the room temperature is a critical factor that has a very strong effect on the EDA signal. A
high room temperature leads to a faster increase in both EDA components (i.e., due to increased
sweat production). Hence, according to existing literature, we kept the room temperature at
about 22 degree Celsius [47].
•
Attention towards physiological and psychological condition: Different physiological as well
as psychological conditions (e.g., tiredness, digestion) are affecting the EDA signal. Since it
is impossible to have participants with the same physiological and psychological condition,
attention should be paid that EDA measurements do not directly follow strongly perceptible
sensations (e.g., hunger).
•
Signal decomposition: The accurate decomposition of the tonic (i.e., SCL) and phasic (i.e., SCR)
component from a raw EDA signal created a vast body of research in this context [
45
]. Since
the two EDA components are located in sensitive frequencies, it is of importance to ensure
that respective methods for analysis are capable of working with fine-grained frequency ranges
(e.g., >.05
µ
s as amplitude threshold for SCR detection as recommended in literature) [
77
].
Therefore, the application of further robust methods for EDA analysis as proposed in literature
is recommended. However, for gaining first experiences (e.g., ambulatory setting) and in the
context of the feasibility study, the used sensor (i.e., EdaMove 3) and related software (i.e.,
DataAnalyzer) seem to be appropriate.
•
Signal transformation: Each individual has a different skin conductivity level depending on
various factors (see Section 2.1). As a result, despite the similar setting, significant differences in
the baseline measurement as well as SCR amplitudes may occur between individuals. For this
reason, obtained EDA results should be standardized. Established methods are log or square
root transformation fostering the difference comparisons between individuals [
47
]. Moreover,
physiological factors (e.g., skin thickness) as well as potential disruptive factors (e.g., Non-SCR)
can be disregarded with specialized transformations.
•
Consideration of more factors: The measurement of the EDA allows for interpretation about
physiological as well as psychological arousal in the presence of stimulus. For many research
purposes (e.g., neural reactions on short-term events), the analysis of the EDA components is
adequate. However, in our context, the sole measurement of the EDA allowed only for limited
interpretation. With the tonic component SCL, we were able to show that the comprehension of
process models poses demands towards cognitive efforts. Regarding the phasic component SCR,
we observed in the feasibility study a higher number of SCR peaks during the comprehension of
the complex process models, but we can only make assumptions (e.g., they may be due to decision
making) regarding their appearance. Therefore, with the addition of further measurements,
better interpretation about the EDA can be assumed. For example, with sensors recording eye
movements, the appearance of SCR peaks can be associated with the gaze of an individual at the
time of a peak.
4. Conclusion and Future Work
This paper presented the first insights about the applicability of measuring the EDA in the
context of process model comprehension. In the scope of the research question, the appropriateness
of measuring the EDA with a specific EDA and physical activity sensor (i.e., EdaMove 3) during the
comprehension of process models was evaluated. Therefore, a preliminary test run and a feasibility
Sensors 2020,xx, 5 18 of 22
study were conducted. The small scale test run was conducted to familiarize ourselves with the
measurement and the analysis of the EDA, and, hence, to obtain the first experiences and lessons
learned. In the feasibility study, n = 9 participants needed to comprehend two differently complex
BPMN 2.0 process models. The results from the feasibility study presented general variations in the
tonic SCL during the comprehension of both process models. Moreover, the complex process model
caused an average higher number of phasic SCR peaks compared to the easy model. Consequently,
participants were confronted with a significantly higher cognitive load (i.e., level of arousal) during
the comprehension of the complex process model. Hence, the number of phasic SCR peaks was
identified in the feasibility study as a significant measure for the determination of distinctions in
the comprehension of process models with varying model complexity. As the first work evaluating
the applicability of measuring the EDA during process model comprehension, this paper made a
contribution to respective research as well as to our existing conceptual framework (see Section 2.2),
which applies measurement methods and theories from cognitive neuroscience and psychology in
order to foster the comprehension of process models towards a neuro-centric perspective [
57
]. We
demonstrated that the measurement of the EDA relying on a smart sensor can be an appropriate
method, especially from a cognitive point of view, to foster our understanding of working with
process models. However, the sole measurement of the EDA is not sufficient to be able to derive
concrete interpretations regarding cognitive processes (e.g., decision making, reasoning) during model
comprehension. Additional psychological or physiological factors have to be taken into account with
the application of further technologies (e.g., heart rate sensor). Therefore, we are currently preparing
another study, in which the EDA is measured simultaneously with recorded eye movements. This
will allow for a better interpretation of especially the phasic SCR peaks, as these can be correlated
with eye movements (e.g., gaze on a stimulus shortly before a SCR peak). A correlation of the
EDA components (i.e., tonic SCL and phasic SCR) with differently complex process model will be
investigated in more detail in the same study. Moreover, although obtained results looked promising,
the conducted feasibility study was confronted with limitations regarding EDA signal decomposition
(e.g., filter frequencies for EDA signal decomposition, see Section 3.3) during the analysis of the
obtained EDA data. Hence, the disclosed limitations need to be addressed carefully in future work. For
this reason, as these limitations are crucial in EDA signal analysis, we strive to analyze the obtained
EDA data with further robust techniques and methods (e.g., SCR peak detection with an amplitude
threshold of >.05
µ
s) from literature, enabling a more rigor conception about the effects in the EDA and
respective components during the comprehension of process models [
45
]. Moreover, the consideration
of further psychophysiological aspects (e.g., heart rate) and related technologies will be subject of
future work. Finally, such psychophysiological data is well suited in this context for further analyses,
such as pattern recognition and machine learning (i.e., linear discriminant analysis) [
78
], and in order
to support our understanding of working with process models. Moreover, such further analyses
allow for the identification of new insights (e.g., objective classification of stakeholders regarding the
comprehension of process models based on psychophysiological measures and individual-related
characteristics) enabling a better support (e.g., comprehension guidelines, tool-assistance) in terms of
process model comprehension in the future.
Author Contributions:
Conceptualization, M.W., R.P., T.P., and M.R.; Methodology, M.W. and T.P.; Software,
M.W.; Validation, M.W., R.P., and T.P.; Formal Analysis, M.W. and T.P.; Investigation, M.W.; Resources, M.W.;
Data Curation, M.W.; Writing–Original Draft Preparation, M.W., R.P., T.P., and M.R.; Writing–Review and Editing,
M.W., R.P., T.P., and M.R.; Visualization, M.W.; Supervision, R.P. and M.R.; All authors have read and agreed to
the published version of the manuscript.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.
Abbreviations
Sensors 2020,xx, 5 19 of 22
The following abbreviations are used in this manuscript:
BPMN Business Process Model and Notation
DC Direct Current
EDA Electrodermal Activity
EEG Electroencephalography
GAPED Geneva Affective Picture Database
SCL Skin Conductance Level
SCR Skin Conductance Response
References
1.
Polyvyanyy, A.; Smirnov, S.; Weske, M. Business process model abstraction. In Handbook on Business Process
Management 1; Springer, Berlin, Germany, 2015; pp. 147–165.
2.
Fan, S.; Hua, Z.; Storey, V.C.; Zhao, J.L. A process ontology based approach to easing semantic ambiguity
in business process modeling. Data & Knowledge Engineering 2016,102, 57–77.
3.
Hammer, M. What is business process management? In Handbook on Business Process Management 1.;
Springer, Berlin, Germany, 2015; pp. 3–16.
4.
Narendra, T.; Agarwal, P.; Gupta, M.; Dechu, S. Counterfactual Reasoning for Process Optimization
Using Structural Causal Models. Proceedings of the 17th International Conference on Business Process
Management (BPM) Forum, Vienna, Austria, 2019, pp. 91–106.
5.
Corradini, F.; Fornari, F.; Polini, A.; Re, B.; Tiezzi, F. A formal approach to modeling and verification of
business process collaborations. Science of Computer Programming 2018,166, 35–70.
6.
Zimoch, M.; Mohring, T.; Pryss, R.; Probst, T.; Schlee, W.; Reichert, M. Using Insights from Cognitive
Neuroscience to Investigate the Effects of Event-Driven Process Chains on Process Model Comprehension.
Proceedings of the 1st International Conference on Cognitive Business Process Management (CBPM),
Barcelona, Spain, 2017, pp. 446–459.
7.
Sánchez-González, L.; García, F.; Ruiz, F.; Piattini, M. A case study about the improvement of business
process models driven by indicators. Software & Systems Modeling 2017,16, 759–788.
8.
Figl, K. Comprehension of Procedural Visual Business Process Models. Business & Information Systems
Engineering 2017,59, 41–67.
9.
Reijers, H.A.; Freytag, T.; Mendling, J.; Eckleder, A. Syntax highlighting in business process models.
Decision Support Systems 2011,51, 339–349.
10.
Milani, F.; Dumas, M.; Matuleviˇcius, R.; Ahmed, N.; Kasela, S. Criteria and heuristics for business process
model decomposition. Business & Information Systems Engineering 2015,58, 7–17.
11.
Mendling, J.; Recker, J.; Reijers, H.A. On the usage of labels and icons in business process modeling.
International Journal of Information System Modeling and Design 2010,1, 40–58.
12.
Kummer, T.; Recker, J.; Mendling, J. Enhancing understandability of process models through
cultural-dependent color adjustments. Decision Support Systems 2016,87, 1–12.
13.
Figl, K.; Mendling, J.; Strembeck, M. The influence of notational deficiencies on process model
comprehension. Journal of the Association for Information Systems 2013,14.
14.
Schrepfer, M.; Wolf, J.; Mendling, J.; Reijers, H.A. The impact of secondary notation on process model
understanding. Proceedings of the 2nd IFIP Working Conference on The Practice of Enterprise Modeling
(POEM), Stockholm, Sweden, 2009, pp. 161–175.
15.
Dikici, A.; Turetken, O.; Demirors, O. Factors influencing the understandability of process models: A
systematic literature review. Information and Software Technology 2018,93, 112–129.
16.
Shahzad, K.; Elias, M.; Johannesson, P. Requirements for a business process model repository: A
stakeholders’ perspective. Proceedings of the International Conference on Business Information Systems
(BIS), Berlin, Germany, 2010, pp. 158–170.
17.
Zimoch, M.; Pryss, R.; Schobel, J.; Reichert, M. Eye tracking experiments on process model comprehension:
lessons learned. Proceedings of the 18th International Workshop on Business Process Modeling,
Development and Support (BPMDS), Essen, Germany, 2017, pp. 153–168.
18.
Mendling, J.; Strembeck, M.; Recker, J. Factors of process model comprehension-findings from a series of
experiments. Decision Support Systems Apr. 2012,53, 195–206.
Sensors 2020,xx, 5 20 of 22
19.
Figl, K.; Recker, J. Exploring cognitive style and task-specific preferences for process representations.
Requirements Engineering 2014,21, 63–85.
20.
Recker, J.; Reijers, H.A.; van de Wouw, S.G. Process model comprehension: the effects of cognitive abilities,
learning style, and strategy. Communications of the Association for Information Systems 2014,34, 9.
21.
Recker, J. Empirical investigation of the usefulness of gateway constructs in process models. European
Journal of Information Systems 2013,22, 673–689.
22.
Turetken, O.; Dikici, A.; Vanderfeesten, I.; Rompen, T.; Demirors, O. The Influence of Using Collapsed
Sub-processes and Groups on the Understandability of Business Process Models. Business & Information
Systems Engineering 2019,62, 121–141.
23.
Kang, H.R.; Yang, H.D.; Rowley, C. Factors in team effectiveness: Cognitive and demographic similarities
of software development team members. Human Relations 2006,59, 1681–1710.
24.
Zimoch, M.; Pryss, R.; Probst, T.; Schlee, W.; Reichert, M. The Repercussions of Business Process Modeling
Notations on Mental Load and Mental Effort. Proceedings of the 11th International Workshop on Social
and Human Aspects of Business Process Management (BPMS), Sydney, Australia, 2018, pp. 133–145.
25.
Chen, T.; Wang, W.; Indulska, M.; Sadiq, S. Business process and rule integration approaches-an empirical
analysis. Proceedings of the 16th International Conference on Business Process Management (BPM),
Sydney, Australia, 2018, pp. 37–52.
26.
Turetken, O.; Vanderfeesten, I.; Claes, J. Cognitive Style and Business Process Model Understanding.
Proceedings of the 29th International Conference on Advanced Information Systems Engineering (CAiSE)
Workshops, Essen, Germany; Metzger, A.; Persson, A., Eds., 2017, pp. 72–84.
27.
Razavian, M.; Turetken, O.; Vanderfeesten, I. When cognitive biases lead to business process management
issues. Proceedings of the 14th International Conference on Business Process Management (BPM)
Workshops, Rio de Janeiro, Brazil, 2016, pp. 147–156.
28.
Mendling, J.; Recker, J.; Reijers, H.A.; Leopold, H. An empirical review of the connection between model
viewer characteristics and the comprehension of conceptual process models. Information Systems Frontiers
2019,21, 1111–1135.
29.
Zimoch, M.; Pryss, R.; Probst, T.; Schlee, W.; Layher, G.; Neumann, H.; Reichert, M. Evaluating the
Comprehensibility of Graphical Business Process Models–An Eye Tracking Study. Proceedings of the 19th
European Conference on Eye Movements (ECEM), Wuppertal, Germany, 2017.
30.
Zimoch, M.; Pryss, R.; Layher, G.; Neumann, H.; Probst, T.; Schlee, W.; Reichert, M. Utilizing the capabilities
offered by eye-tracking to foster novices’ comprehension of business process models. Proceedings of the
2nd International Conference on Cognitive Computing (ICCC), Seattle, USA, 2018, pp. 155–163.
31.
Petrusel, R.; Mendling, J.; Reijers, H.A. How visual cognition influences process model comprehension.
Decision Support Systems 2017,96, 1–16.
32.
Wang, W.; Indulska, M.; Sadiq, S.; Weber, B. Effect of linked rules on business process model understanding.
Proceedings of the 15th International Conference on Business Process Management (BPM), Barcelona,
Spain, 2017, pp. 200–215.
33.
Tallon, M.; Winter, M.; Pryss, R.; Rakoczy, K.; Reichert, M.; Greenlee, M.W.; Frick, U. Comprehension of
business process models: Insight into cognitive strategies via eye tracking. Expert Systems with Applications
2019,136, 145–158.
34.
Winter, M.; Pryss, R.; Probst, T.; Reichert, M. Learning to Read by Learning to Write: Evaluation of a
Serious Game to Foster Business Process Model Comprehension. JMIR Serious Games 2020,8.
35.
Ferhat, O.; Vilariño, F. Low cost eye tracking: The current panorama. Computational Intelligence and
Neuroscience 2016,2016.
36.
Zhou, Z.; Liao, H.; Gu, B.; Huq, K.M.S.; Mumtaz, S.; Rodriguez, J. Robust mobile crowd sensing: When
deep learning meets edge computing. IEEE Network 2018,32, 54–60.
37.
Coppetti, T.; Brauchlin, A.; Müggler, S.; Attinger-Toller, A.; Templin, C.; Schönrath, F.; Hellermann, J.;
Lüscher, T.F.; Biaggi, P.; Wyss, C.A. Accuracy of smartphone apps for heart rate measurement. European
Journal of Preventive Cardiology 2017,24, 1287–1293.
38.
Ganster, D.C.; Crain, T.L.; Brossoit, R.M. Physiological measurement in the organizational sciences: A
review and recommendations for future use. Annual Review of Organizational Psychology and Organizational
Behavior 2018,5, 267–293.
Sensors 2020,xx, 5 21 of 22
39.
Wichary, S.; Mata, R.; Rieskamp, J. Probabilistic inferences under emotional stress: how arousal affects
decision processes. Journal of Behavioral Decision Making 2016,29, 525–538.
40.
Kuppens, P.; Tuerlinckx, F.; Yik, M.; Koval, P.; Coosemans, J.; Zeng, K.J.; Russell, J.A. The relation between
valence and arousal in subjective experience varies with personality and culture. Journal of Personality
2017
,
85, 530–542.
41.
Kusserow, M.; Amft, O.; Tröster, G. Modeling arousal phases in daily living using wearable sensors. IEEE
Transactions on Affective Computing 2012,4, 93–105.
42.
Meißner, M.; Oll, J. The promise of eye-tracking methodology in organizational research: A taxonomy,
review, and future avenues. Organizational Research Methods 2019,22, 590–617.
43.
Fu, R.; Han, M.; Wang, F.; Shi, P. Intentions Recognition of EEG Signals with High Arousal Degree for
Complex Task. Journal of Medical Systems 2020,44, 1–12.
44.
Picard, R.W.; Fedor, S.; Ayzenberg, Y. Multiple arousal theory and daily-life electrodermal activity
asymmetry. Emotion Review 2016,8, 62–75.
45.
Posada-Quintero, H.F.; Chon, K.H. Innovations in Electrodermal Activity Data Collection and Signal
Processing: A Systematic Review. Sensors 2020,20, 479.
46.
Affanni, A. Wireless Sensors System for Stress Detection by Means of ECG and EDA Acquisition. Sensors
2020,20.
47.
Braithwaite, J.J.; Watson, D.G.; Jones, R.; Rowe, M. A guide for analysing electrodermal activity (EDA) &
skin conductance responses (SCRs) for psychological experiments. Psychophysiology 2013,49, 1017–1034.
48.
Ayaz, H. Electrodermal Activity in Ambulatory Settings: A Narrative Review of Literature. Advances in
Neuroergonomics and Cognitive Engineering 2019,953, 91–102.
49.
Shukla, J.; Barreda-Angeles, M.; Oliver, J.; Nandi, G.; Puig, D. Feature Extraction and Selection for Emotion
Recognition from Electrodermal Activity. IEEE Transactions on Affective Computing 2019.
50.
Al Machot, F.; Elmachot, A.; Ali, M.; Al Machot, E.; Kyamakya, K. A deep-learning model for
subject-independent human emotion recognition using electrodermal activity sensors. Sensors
2019
,
19, 1659–1673.
51.
Shaffer, F.; Combatalade, D.; Peper, E.; Meehan, Z.M. A guide to cleaner electrodermal activity
measurements. Biofeedback 2016,44, 90–100.
52.
Sakai, T.; Tamaki, H.; Ota, Y.; Egusa, R.; Inagaki, S.; Kusunoki, F.; Sugimoto, M.; Mizoguchi, H.; others.
Eda-Based Estimation Of Visual Attention By Observation Of Eye Blink Frequency. International Journal on
Smart Sensing and Intelligent Systems 2017,10, 296–307.
53.
Dawson, M.E.; Schell, A.M.; Filion, D.L. The electrodermal system. In Handbook of Psychophysiology, 4th ed.;
Cambridge University Press, 2017; pp. 217–243.
54.
Mason, L.; Scrimin, S.; Tornatora, M.C.; Zaccoletti, S. Emotional reactivity and comprehension of multiple
online texts. Learning and Individual Differences 2017,58, 10–21.
55.
Chung, S.; Cheon, J.; Lee, K.W. Emotion and multimedia learning: an investigation of the effects of valence
and arousal on different modalities in an instructional animation. Instructional Science 2015,43, 545–559.
56.
Scrimin, S.; Mason, L. Does mood influence text processing and comprehension? Evidence from an
eye-movement study. British Journal of Educational Psychology 2015,85, 387–406.
57.
Zimoch, M.; Pryss, R.; Probst, T.; Schlee, W.; Reichert, M. Cognitive insights into business process model
comprehension: Preliminary results for experienced and inexperienced individuals. Proceedings of the
18th International Conference on Business Process Modeling, Development and Support (BPMDS), Essen,
Germany, 2017, pp. 137–152.
58.
Movisens. EdaMove 3. https://www.movisens.com/de/produkte/eda-und-aktivitaetssensor-edamove-3/,
last accessed on 11/06/20202.
59.
Beiwinkel, T.; Hey, S.; Bock, O.; Rössler, W. Supportive mental health self-monitoring among smartphone
users with psychological distress: protocol for a fully mobile randomized controlled trial. Frontiers in
Public Health 2017,5, 249.
60.
Santangelo, P.S.; Holtmann, J.; Hosoya, G.; Bohus, M.; Kockler, T.D.; Koudela-Hamila, S.; Eid, M.;
Ebner-Priemer, U.W. Within-and Between-Persons Effects of Self-Esteem and Affective State as Antecedents
and Consequences of Dysfunctional Behaviors in the Everyday Lives of Patients With Borderline
Personality Disorder. Clinical Psychological Science 2020,8, 428–449.
61. Boucsein, W. Electrodermal activity, 2 ed.; Springer, 2012.
Sensors 2020,xx, 5 22 of 22
62.
Dan-Glauser, E.S.; Scherer, K.R. The Geneva affective picture database (GAPED): a new 730-picture
database focusing on valence and normative significance. Behavior Research Methods 2011,43, 468–477.
63.
Matamala, A.; Soler-Vilageliu, O.; Iturregui-Gallardo, G.; Jankowska, A.; Méndez-Ulrich, J.L.; Serrano,
A. Electrodermal activity as a measure of emotions in media accessibility research: methodological
considerations. Journal of Specialised Translation 2020,33, 129–151.
64.
Radin, D.I. Electrodermal presentiments of future emotions. Journal of Scientific Exploration
2004
,
18, 253–273.
65.
OMG, Object Management Group Specification. Business Process Modeling & Notation 2.0.
https://www.bpmn.org, last accessed on 11/06/2020.
66.
Mendling, J. Metrics for process models: empirical foundations of verification, error prediction, and guidelines for
correctness, 6 ed.; Springer, 2008.
67.
Wohlin, C.; Runeson, P.; Höst, M.; Ohlsson, M.C.; Regnell, B.; Wesslen, A. Experimentation in Software
Engineering - An Introduction, 1 ed.; Kluwer, 2012.
68.
Larmuseau, C.; Vanneste, P.; Cornelis, J.; Desmet, P.; Depaepe, F. Combining physiological data and
subjective measurements to investigate cognitive load during complex learning. Frontline Learning Research
2019,7, 57–74.
69.
Mandryk, R.L.; Atkins, M.S. A fuzzy physiological approach for continuously modeling emotion during
interaction with play technologies. International Journal of Human-Computer Studies 2007,65, 329–347.
70.
Boucsein, W.; Fowles, D.C.; Grimnes, S.; Ben-Shakhar, G.; Roth, W.T.; Dawson, M.E.; Filion, D.L. Publication
recommendations for electrodermal measurements. Psychophysiology 2012,49, 1017–1034.
71.
Posada-Quintero, H.F.; Dimitrov, T.; Moutran, A.; Park, S.; Chon, K.H. Analysis of reproducibility of
noninvasive measures of sympathetic autonomic control based on electrodermal activity and heart rate
variability. IEEE Access 2019,7, 22523–22531.
72.
Momin, A.; Shahu, A.; Sanyal, S.; Chakraborty, P. Electrodermal activity and its effectiveness in cognitive
research field. In Cognitive Informatics, Computer Modelling, and Cognitive Science; Elsevier, 2020; Vol. 2, pp.
149–177.
73.
Fritz, T.; Begel, A.; Müller, S.C.; Yigit-Elliott, S.; Züger, M. Using psycho-physiological measures to assess
task difficulty in software development. Proceedings of the 36th International Conference on Software
Engineering (ICSE), Hyderabad, India, 2014, pp. 402–413.
74.
Liu, Y.; Du, S. Psychological stress level detection based on electrodermal activity. Behavioural Brain
Research 2018,341, 50–53.
75.
Khan, T.H.; Villanueva, I.; Vicioso, P.; Husman, J. Exploring relationships between electrodermal activity,
skin temperature, and performance during. IEEE Frontiers in Education Conference (FIE), 2019, pp. 1–5.
76.
Kappeler-Setz, C.; Gravenhorst, F.; Schumm, J.; Arnrich, B.; Tröster, G. Towards long term monitoring of
electrodermal activity in daily life. Personal and Ubiquitous Computing 2013,17, 261–271.
77.
Posada-Quintero, H.F.; Florian, J.P.; Orjuela-Cañón, A.D.; Aljama-Corrales, T.; Charleston-Villalobos, S.;
Chon, K.H. Power spectral density analysis of electrodermal activity for sympathetic function assessment.
Annals of biomedical engineering 2016,44, 3124–3135.
78.
Pryss, R.; Schlee, W.; Hoppenstedt, B.; Reichert, M.; Spiliopoulou, M.; Langguth, B.; Breitmayer, M.;
Probst, T. Applying Machine Learning to Daily-Life Data From the TrackYourTinnitus Mobile Health
Crowdsensing Platform to Predict the Mobile Operating System Used With High Accuracy: Longitudinal
Observational Study. Journal of Medical Internet Research 2020,22.
©
2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
