A Methodology for Determination of Performance Measures
Thresholds for Business Process
Mariem Kchaou, Wiem Khlif and Faiez Gargouri
Mir@cl Laboratory, University of Sfax, Sfax, Tunisia
Keywords: Business Process, Performance Measures, Fuzzy Logic, Thresholds, Characteristics Related to BPMN
Elements, Characteristics Related to the Actor.
Abstract: Business process performance is vital for organizations which aim to produce a high performance model. In
the literature, performance of the business process can be evaluated through formal verifications, simulation,
or a set of measures. In this paper, we adopt measures-based assessment to evaluate the performance of
business process models, modelled with Business Process Modelling and Notation (BPMN), in terms of the
characteristics related to BPMN elements (i.e. time behaviour, cost ) and characteristics related to the actor
(ie. availability, suitability and its cost). We propose a methodology based on fuzzy logic which apply
performance measures to assess these characteristic’s levels. In addition, it expresses the problem of defining
threshold based on a set of BPMN models²s. Furthermore, our methodology evaluates the performance of
business process models based on fuzzy logic. The efficiency of the proposed methodology is illustrated
through a case study and a tool that fully support the developed system.
1 INTRODUCTION
Performance is necessary step for enterprises, seeking
to improve their business process (BP). Evidently, BP
performance aims to reduce time, cost and to indicate
whether the company goals are achieved or not.
In the literature, BP model performance
assessment shows two trends of approaches: those
centred on the application of formal verification
methods (Kluza and Nalepa, 2019) or those based on
the use of a set of performance measures calculated
on the BP model (Lanz et al., 2016) (Khlif et al.,
2019) (Kchaou et al., 2019).
Formal methods are used to verify performance
properties like measurement process and feedback
process (Kluza and Nalepa, 2019). However, their
application stills delayed by their time and cost. In
addition, they are not able to analyse the model
performance such as its time behaviour and cost of
BPMN elements; and also availability, suitability and
cost of the actor. These characteristics influence the
performance of the BP.
In addition, several authors adopts a qualitative
assessment of BP models by proposing a set of
performance measures that are applied either on the
BP model (e.g. (Kis et al., 2017) (Khlif et al., 2019)),
or the simulated BP model (Heinrich, 2013)
(D'Ambrogio et al., 2016). These measures are
exploited to assess several quality characteristicS
(Razzaq et al., 2018) (Gonzalez-Lopez and Bustos,
2019) or to predict the BP performance (case of
simulated model assessment) (Heinrich, 2013)
(D'Ambrogio et al., 2016).
Since the diversity of measures, several
researchers proposed frameworks to evaluate the
performance of a business process model e.g. (Wynn
et al., 2013), (Kis et al., 2017) (Khlif et al., 2019).
However, there is no consensus about threshold
values of performance measures which are required
to interpret/evaluate a BP model’s performance.
This paper overcomes the problem of threshold
identification based on fuzzy logic methodology
which asses the BP performance in terms of
characteristics such as the time behaviour and cost of
BPMN elements; availability, suitability and cost of
the actor. These characteristics are crucial to improve
the business process model.
The proposed methodology proceeds in two
phases: threshold identification and fuzzy logic
application. First, it uses data mining to define the
decision tree, which identify approximate thresholds
for each performance measure. These thresholds
allow the designer to interpret the characteristic of
144
Kchaou, M., Khlif, W. and Gargouri, F.
A Methodology for Determination of Performance Measures Thresholds for Business Process.
DOI: 10.5220/0009397801440157
In Proceedings of the 15th International Conference on Evaluation of Novel Approaches to Software Engineering (ENASE 2020), pages 144-157
ISBN: 978-989-758-421-3
Copyright
c
 2020 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

Business Process Modelling Notation (BPMN)
elements (i.e., time behaviour, and cost levels) and
those related to the actor (i.e., availability, suitability
or cost levels). To this end, we used a database
intituled "Business Process Database". We collect
100 business processes of organizations operating in
different sectors, and then we annotate them by
temporal and semantic information. Our database is
available
at:
https://sites.google.com/site/kchaoumariemsi/resources.
The approximate thresholds produced in the first
phase are considered as the input of the second phase.
This phase uses the fuzzy logic (Zadeh, 2008) in order
to obtain precise thresholds values.
The proposed methodology is developed in a tool
that help to evaluate the performance of BPMN
models in terms of time behaviour and cost of BPMN
elements; and availability, suitability and cost of the
actor. To illustrate the efficiency of our performance
tool, we rely on two types of experimental evaluation.
The former is accomplished with students while the
second is done through the proposed tool.
In summary, this paper presents two
contributions: the first one expresses the imprecise
thresholds determination for performance measures
in terms of time behaviour and cost of BPMN
elements; and availability, suitability and cost of the
actor. The second one handle the imprecise nature of
the identified thresholds by applying fuzzy logic.
The remainder of the paper is organized as
follows: Section 2 summarizes related work. In
Section 3, we present the proposed methodology.
Section 4 expresses how we apply fuzzy logic to
support the imprecise thresholds. Section 5 illustrate
the developed tool of BP model performance
assessment and evaluate it based on two types of
experiments. Section 6 identifies threats to the
validity of our methodology. Finally, section 7
summarizes the presented work and outlines its
extensions.
2 RELATED WORK
In this section, we overview works on the BP
performance measures. These works are divided into
two categories: measures related to the actor
characteristics and those related to BPMN elements
characteristics.
It is to note that the presented measures below are
those having formula that allow calculating the value
of each one in a BPMN model. Based on this
criterion, we retain all of them for the determination
of their thresholds.
2.1 Measures Related to the Actor
Characteristics
In (Khlif et al., 2019) (Kchaou et al., 2019), to
evaluate the performance of an actor, the authors
propose measures related to the actor characteristics
such as availability, suitability and cost.
Availability is the capability of the actor to be able
to perform the activity in the required unit of time.
Suitability expresses the skills that cover his
qualification, expertise, social competence, skills,
motivation and performance ability. The cost is
expressed as a price or monetary value.
The following measures assess the availability
and suitability of the actor:
 Planned Production time of an Actor to
perform an Activity (PPT
Act
(A)): is calculated
by subtracting the Actor’s BReaks
(unproductive time where the actor is
scheduled not to work) from Shift time (a
period where an actor is scheduled to perform
an Activity).
 Working Time spent by an Actor to perform an
Activity (WT
Act
(A)): is simply calculated by the
difference between the Planned Production
Time and Stop Time (the time where the actor
was intended to work but was not due to
unplanned stops or planned stops).
 Total Working Time spent by an Actor in a
Lane per Day (TWTDay
Act
(L)) : the sum of
working time spent, in a day, by an actor in the
corresponding lane.
 Total Working Time spent by an Actor in the
whole Process per Day (TWTDay
Act
(P)) : the
sum of working time spent by an actor in all
lanes in the process.
 Performance of an Actor per Day (PerDay
Act
):
compares the working Time spent by an actor
per day to the Ideal Cycle Time which is
defined as the theoretical minimum time to
perform an activity by an actor.
 Availability of an Actor in a Day (AVDay
Act
):
is calculated as the ratio of Working Time spent
by an actor to Planned Production Time.
 Ratio of Defected Activities by an Actor per day
(RDA
Act
): is calculated by the Total Number of
Defected Activities performed by an actor
divided by the Total number of Activities
performed by the same actor.
 Ratio of Good Activities performed by an Actor
(RGA
Act
): is calculated by the Total Number of
Good Activities realized by an actor in a day
divided by the Total number of Activities
performed by the same actor in one day.
A Methodology for Determination of Performance Measures Thresholds for Business Process
145

In addition, several measures are proposed in (Khlif
et al., 2019) (Kchaou et al., 2019) to assess the cost of
an actor such as Cost of an actor in a Lane per Day
(CosDay
act
(L)) which is calculated by the product of
the total working time spent by an Actor in a Lane per
Day (TWTDay
Act
(L)) and its actual Labour Costs per
Hour (LCH
Act
), Cost of an actor in a Pool per Day
(CosDay
Act
(P)) which is determined by the product of
the total working time spent by an Actor in a Pool per
Day (TWTDay
Act
(P)) and its actual Labour Costs per
Hour (LCH
Act
).
2.2 Measures Related to BPMN
Elements Characteristics
Time behaviour and cost are the characteristics of
BPMN elements to evaluate the performance
efficiency (Heinrich and Paech, 2010).
Time behaviour is defined as the appropriate
transport time between different BPMN elements and
processing times when executed; while cost expresses
the price or monetary value related to BPMN elements.
On the one hand, a set of measures are proposed
in (Khlif et al., 2019) (Kchaou et al., 2019) to assess
the time behaviour of BPMN elements such as
Gateway Duration (GD (Gateway) which represents
the duration of a gateway. In addition, (Lanz et al.,
2016) propose other temporal measures such as
Activity/Process Duration (AD) which is calculated
by the difference between the end time of the activity
(Process) and the start time.
On the other hand, (Khlif et al., 2019) (Kchaou et
al., 2019) proposed a set of measures to evaluate the
cost of BPMN elements such as Cost of an Activity
realized by an actor (CA
Act
) which is calculated by the
product of the actor actual Labour Costs per Hour and
the working time spent by an Actor to perform an
Activity; and Cost of a Gateway (CosGat(Gatway))
which represents the product of the gateway duration
and the actor’s actual Labour Costs per Hour (LCH
Act
).
Table 5 and 7 show respectively the usability of
these measures to assess the actor characteristics and
BPMN element characteristics. However, to our
knowledge, there is no works that focus on the
determination of measures thresholds values.
3 DESIGN METHODOLOGY FOR
THRESHOLDS
DETERMINATION
Figure 1 depicts our methodology for threshold
determination to assess the cost and time of BPMN
elements and evaluate the suitability, availability and
cost of the actor.
Our design methodology followed two major
phases: “Analyze Data” and “Validate Data”.
The activities of the “Analyze data” are organized
essentially in three stages: the first one collects data
based on a set of business process models annotated
by temporal constraints and semantic information
(cost and organizational aspects). The second step
prepares data to test the database and the third one
apply data mining technique to build decision trees.
The second phase "Validate Data" is composed of two
activities: Training Database based Validation and
Test Database based Validation.
3.1 Analyze Data
The Analyze data phase goes through three major
stages: 1) Collect a set of BPMN models that we
annotated by semantic and temporal information, 2)
Prepare these models through creating matrices
related to actors and to BPMN elements to evaluate
their characteristics and 3) Apply Data mining to
build decision trees using WEKA system. The latter
is based on algorithms that construct decision trees.
3.1.1 Collect Database
In the first step, we collect 100 BPMN models having
small/ medium size, and belonging to different
organizations such as banks, healthcare, institutions,
commercial enterprises, etc. Then, we annotate them
by semantic information that covers the cost,
organizational aspect, and temporal constraints
related to BPMN elements and the actor. For more
details, reader can refer to (Kchaou et al., 2019). This
information are used to evaluate the actor and BPMN
elements characteristics.
Next, we examined business processes with
experts according to measures values related to each
characteristics associated to the actor and to BPMN
elements. The objective is to organize them according
to the level of each characteristic related to the actor
and BPMN elements.
To end this purpose, we organized ourselves into
two groups. First, each one examine 50 processes in
term of characteristics related to the actor and to the
BPMN elements. Then, we verify the cross-
validation process among the two groups. Finally, to
assess business processes based on the actor
characteristics, we organized the "Business Process
Database" into two levels of suitability (having the
best skills and having low skills), two levels of
availability (always available and rarely available)
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
146

Figure 1: Design methodology for thresholds determination.
and three levels of the cost (expensive, acceptable
and cheap).
To evaluate the process in terms of characteristics
BPMN elements, we classified the "Business Process
Database" in three levels of time behaviour (minimal,
normal and maximal) and the three levels of cost
(expensive, acceptable and cheap).
3.1.2 Prepare Data
In order to prepare data for the next stage, we take as
input performance measures values and the level of
each characteristics related to actors and to BPMN
elements to produce nine matrices based on the
"Business Process Database". Three matrices are
devoted to the actor in order to measure his
availability, suitability and cost; while the rest is
associated to the BPMN elements (activity, gateway
and sequence flow) to evaluate their time behavior
and cost.
Each row in each matrix expresses the actor
(respectively BPMN element); and each column
depicts a performance measure used to assess the
availability, suitability and cost of the actor
(respectively time behaviour and cost of BPMN
elements). The corresponding case representing the
intersection of row and column details the values of
these performance measures calculated for a specific
actor (respectively BPMN elements).
The last column of each matrix represents the
level of each actor characteristic (respectively BPMN
element). For example, the last column of each matrix
associated to the actor represents the level of his
availability (i.e., actor is always available and rarely
available), suitability (i.e., having the best skills and
having low skills) and cost (i.e., expensive,
acceptable and cheap).
The elaborated matrices are used to create two
sub-datasets: one for learning "Training Dataset"
which comprises 70% of the "Business Process
Database" and one for testing needs "Test Dataset"
which includes the rest of the "Business Process
Database". The percentage choice is justified by the
fact that the "Training Dataset" is the one on which
we train and fit our model to adjust thresholds.
Whereas "Test Dataset" is used only to assess the BP
performance.
A Methodology for Determination of Performance Measures Thresholds for Business Process
147

3.1.3 Data Mining
To extract thresholds for performance measures from
the "Business Process Database" and evaluate the
performance of a business process model (BPM), we
used in the first stage decision trees and in the second
stage decision rules.
A decision tree has a root node, intermediate and
terminal nodes. The root node represents the
"Business Process Database" which is divided into
two or more homogeneous sets. Terminal node
represent the level of each BPMN element
characteristic (time behavior and cost) and each actor
characteristic (the availability, suitability and the
cost). The transitions from the root node to a terminal
node are based on the values of performance
measures. For each node, the value of performance
measure that maximizes the homogeneity of child
nodes is chosen. Node homogeneity is attained if all
the BPs of this node belong to the same level (e.g., all
the BP of a node are expensive, in the case of cost).
A homogeneous node is usually a leaf node. In the
case of BPMN element characteristic, a leaf node
represents a class, expressing the level of cost or the
level of time behaviour.
At the same, we apply this interpretation to the
actor characteristics. To create decision trees, we use
the training dataset which contains the values of the
performance measures calculated for a specific actor
(respectively BPMN elements). The required nine
decision trees is classified into three for the actor
characteristics (Availability, suitability and cost) and
six for BPMN elements characteristics (time
behaviour and cost), we used WEKA system (Hall et
al., 2009) which is a collection of machine learning
algorithms for data mining tasks. It contains tools for
data pre-processing, classification, regression,
clustering, association rules, and visualization.
WEKA is based on algorithms (J48, RandomTree,
REPTree, etc.) that construct decision trees. We note
that the J48 algorithm is an implementation of C4.5
algorithm (
Chen et al., 2009). It produces decision tree
classification for a given dataset by recursive division
of the data.
It works with the process of starting from leaves
that overall formed tree and do a backward toward the
root. The RepTree uses the regression tree logic and
creates multiple trees in different iterations. After that
it selects best one from all generated trees. The
Random Tree is a supervised Classifier; it is an
ensemble learning algorithm that generates many
individual learners. It employs a bagging idea to
produce a random set of data for constructing a
decision tree.
In this work, we first apply all of the algorithms,
and then we choose the best one which have a lower
error rate based on the validation phase (Section 3.2).
3.2 Validate Data
In order to evaluate the quality of a prediction model,
we apply various ratios like precision (1), recall (2),
f-measure (3), and global error rate (4). Afterward, we
choose the most popular and best algorithms based on
the values of the used ratios such as J48,
RandomTree, and REPTree.
iesFoundTotalEntit
itiesFoundCorrectEnt
=ecisionPr
(1)
ctEntitiesTotalCorre
itiesFoundCorrectEnt
=callRe
(2)
callRe+ecisionPr
callRe*ecisionPr
*2=mesure_F
(3)
iesTotalEntit
itiesFoundCorrectEnt
1=rRateGlobalErro
_
(4)
3.2.1 Training Database based Validation
Mining
We start by calculating the ratios after testing the
resulting decision trees based on the availability,
suitability and cost trees of the actor, and also based
on time behaviour and cost trees of BPMN elements.
Decision trees are applied on the "Training
Database".
Table 1 expresses that we reached very acceptable
results with REPTree algorithm, for evaluating the
BP model actor characteristics. Regarding the
availability, the values of precision, recall, and F-
measure are 94.5%, 94.1% and 94.2% while the
global error is equal to 5.8%. To evaluate the
suitability, the values of precision, recall, and F-
measure are 76.4%, 76.5% and 76.3% while the error
is equal to 2.3%. In addition, regarding the cost, the
values of precision, recall, and F-measure are 98.6%,
98.5% and 98.5% while the global error rate is 1.4%.
Table 2 shows that we achieved very acceptable
results with REPTree algorithm, for assessing BPMN
elements characteristics. To evaluate each
characteristic, we calculate for each one the values of
precision, recall, and F-measure and the
corresponding errors.
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
148

Table 1: J84 vs RandomTree vs REPTree for decision tree of availability, suitability and cost of the actor using the "Training
Database".
Ratios Availabilit
y
Suitabilit
y
Cost
J48 RandomTree REPTree J48 RandomTree REPTree J48 RandomTree REPTree
Precision 0,815 0,869 0,945 0,748 0,724 0,764 0,972 0,986 0,986
Recall 0,824 0,863 0,941 0,706 0,725 0,765 0,971 0,985 0,985
F-Measure 0,808 0,866 0,942 0,704 0,724 0,763 0,971 0,985 0,985
Global error
rate
0.176 0.137 0.058 0.029 0.027 0.023 0.029 0.014 0.014
Table 2: J84 vs RandomTree vs REPTree for decision tree of time behaviour and cost of each BPMN elemnt using the
"Training Database".
BPMN
elements
Ratios Time behaviou
r
Cost
J48 RandomTree REPTree J48 RandomTree REPTree
activity Precision 0,987 0,987 0,991 0,968 0,964 0,969
Recall 0,986 0,986 0,991 0,968 0,964 0,968
F-Measure 0,986 0,986 0,991 0,968 0,964 0,968
Global error rate 0,013 0,013 0,009 0.031 0.036 0.031
Gateway Precision 0,989 0,989 0,989 0,955 0,932 0,980
Recall 0,988 0,988 0,988 0,955 0,932 0,977
F-Measure 0,988 0,988 0,988 0,955 0,931 0,978
Global error rate 0,011 0,011 0,011 0.045 0.068 0.022
Sequence
Flow
Precision 0,980 0,975 0,980 0,967 0,983 0,983
Recall 0,980 0,975 0,980 0,964 0,982 0,982
F-Measure 0,980 0,975 0,980 0,964 0,982 0,982
Global error rate 0,020 0,025 0,020 0.035 0.017 0.017
Table 3: J84 vs RandomTree vs REPTree for decision tree of availability, suitability and cost of the actor using the "Test
Database".
Ratios Availabilit
y
Suitabilit
y
Cost
J48 RandomTree REPTree J48 RandomTree REPTree J48 RandomTree REPTree
Precision 0,917 0,887 0,917 0,702 0,634 0,870 0,889 0,923 0,965
Recall 0,917 0,875 0,917 0,700 0,633 0,867 0,885 0,923 0,962
F-Measure 0,917 0,879 0,917 0,700 0,627 0,867 0,884 0,923 0,962
Global error
rate
0.083 0.125 0.083 0.030 0.036 0.013 0.115 0.076 0.038
3.2.2 Test Database based Validation
Mining
To evaluate the performance of the proposed decision
tree and select the best algorithm provided by WEKA,
we use the "Test Database", which is extracted from
the "Business Process Database".
Then, we assess the level of each characteristic
related to the actor (the availability, suitability and
cost levels of each actor) and BPMN elements (the
time behaviour and cost levels of each BPMN
elements) by applying each decision tree to all BPs of
the "Test Database". Then, we compare this
evaluation to the assessment already done by experts.
The objective behind is to compare the obtained
decision trees with expert judgment and therefore, to
determine the error rate of our decision trees.
Tables 3 and 4 depict the values of the ratios
presented in section 3.2 for assessing the performance
of the proposed characteristics of decision trees
(availability, suitability and the cost of the actor and
time behavior and cost of BPMN elements).
Table 3 displays that we realized very acceptable
results using the "Test Database" with REPTree
algorithm, for assessing the actor characteristics.
Regarding the availability, the values of precision,
recall, and F-measure are 91.7%, 91.7% and 91.7%
while the global error is equal to 8.3%. To evaluate
the suitability, the values of precision, recall, and F-
measure are 87%, 86.7% and 86.7% while the error is
equal to 1.3%. In addition, regarding the cost, the
values of precision, recall, and F-measure are 96.5%,
96.2% and 96.2% while the global error rate is 3.8%.
In addition, based on Table 4, we deduce that the
attained results with REPTree algorithm are very
acceptable, for assessing BPMN elements
characteristics. To evaluate each characteristic, we
calculate for each one the values of precision, recall,
and F-measure and the corresponding errors.
A Methodology for Determination of Performance Measures Thresholds for Business Process
149

Table 4: J84 vs RandomTree vs REPTree for decision tree of time behaviour and cost of each BPMN element using the "Test
Database".
BPMN
elements
Ratios Time behaviou
r
Cost
J48 RandomTree REPTree J48 RandomTree REPTree
activity Precision 0,987 0,987 0,991 0,968 0,964 0,969
Recall 0,986 0,986 0,991 0,968 0,964 0,968
F-Measure 0,986 0,986 0,991 0,968 0,964 0,968
Global error rate 0,013 0,013 0,009 0.031 0.036 0.031
Gateway Precision 0,989 0,989 0,989 0,955 0,932 0,980
Recall 0,988 0,988 0,988 0,955 0,932 0,977
F-Measure 0,988 0,988 0,988 0,955 0,931 0,978
Global error rate 0,011 0,011 0,011 0.045 0.068 0.022
Sequence
Flow
Precision 0,932 0,979 0,979 0,946 0,946 0,946
Recall 0,932 0,977 0,977 0,946 0,946 0,946
F-Measure 0,931 0,977 0,977 0,946 0,946 0,946
Global error rate 0,068 0.022 0.022 0.054 0.054 0.054
3.3 Discussion
According to the level of each characteristic related
to the actor (availability, suitability and cost levels of
each actor) and BPMN elements (time behaviour and
cost levels of each BPMN element), we used decision
trees to classify BPMN elements and actors extracted
from "Business Process Database". This
classification depend on the values of the used
performance measures.
Furthermore, these decision trees are used to
determine a set of decision rules and performance
measures thresholds to asses the availability,
suitability and cost of each actor (respectively time
behaviour and cost of each BPMN element).
3.3.1 Evaluation of Actor Characteristics
Levels
Table 5 illustrates the thresholds values and their
interpretations which are determined by experts in our
laboratory.
Table 6 depicts an extract of the decision rules
which indicate the performance measures values for
each characteristic level of the actor.
3.3.2 Evaluation of Actor Characteristics
Levels
Table 7 illustrates thresholds and the corresponding
linguistic interpretations, which are determined by the
members of our research team.
Table 8 displays an extract of decision rules that
define the level of each BPMN element
characteristics based on the values of performance
measures
.
Table 5: Identified thresholds values for the evaluation of
the characteristics related to the actor.
Performance
measures
Threshold Linguistic
interpretation
Availability
PPT
Act
(A) PPTAct
(
A
)
< 5 Low
5<=PPTAct
(
A
)
<7 Moderate
7<=PPTAct
(
A
)
<9 Hi
g
h
PPTAct
(
A
)
>= 9 Ver
y
hi
g
h
WT
Act
(A) WT
Act
(
A
)
<3 Low
WT
Act
(
A
)
>=3 Hi
g
h
PerDay
Act
PerDa
y
Act
<78 Low
PerDay
Act
>=78 High
AVDay
Act
AVDa
y
act < 72.5 Low
AVDa
y
act >= 72.5 Hi
g
h
Suitabilit
y
PPT
Act
(A) PPTAct
(
A
)
< 12.5 Low
12.5<=PPTAct
(
A
)
<17.5 Moderate
17.5<=PPTAct
(
A
)
<25 Hi
g
h
PPTAct
(
A
)
>= 25 Ver
y
hi
g
h
WT
Act
(A) WT
Act
(
A
)
<3 Ver
y
low
3<=WT
Act
(A)<10.5 Low
10.5<= WT
Act
(
A
)
<12.5 Moderate
12.5<=WT
Act
(A)<21 High
WT
Act
(
A
)
>=21 Ver
y
hi
g
h
TWTDay
Act
(L) TWTDa
y
Act
(
L
)
<17.5 Low
17.5<=TWTDa
y
Act
(
L
)
<24.5 Moderate
TWTDa
y
Act
(
L
)
>=24.5 Hi
g
h
TWTDay
Act
(P) TWTDa
y
Act
(
P
)
<60 Low
TWTDa
y
Act
(
P
)
>=60 Hi
g
h
PerDay
Act
PerDa
y
Act
<79.2 Low
PerDa
y
Act
>=79.2 Hi
g
h
AVDay
Act
AVDayact < 72.5 Low
AVDa
y
act >= 72.5 Hi
g
h
RGAAct RGAAct<37.5 Low
37.5<= RGAAct<75 Moderate
RGAAct>=75 Hi
g
h
RDAAct RDAAct<58.3 Low
RDAAct>=58.3 Hi
g
h
Cost
CosDayAct(L) CosDa
y
Act
(
L
)
<4.8 Low
4.8<=CosDa
y
Act
(
L
)
<9 Moderate
CosDa
y
Act
(
L
)
>=9 Hi
g
h
CosDayAct(P) CosDa
y
Act
(
P
)
<10.16 Low
CosDa
y
Act
(
P
)
>=10.16 Hi
g
h
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
150

Table 6: Extract of decision rules to assess the level of actor’s characteristics.
Characteristics Rule Decision rules
Availability R1 If
(
AVDa
y
act < 72.5 and PPTAct
(
A
)
< 9 and WTAct
(
A
)
< 3
)
then the actor is rarel
y
available
R2 If (AVDayact < 72.5 and PPTAct(A) >= 9 and PerDayAct >= 78 and TWTDayAct(L) < 15.5) then the actor is always
available
R3 If (AVDayact >= 72.5 and PPTAct(A) < 7 and PPTAct(A) < 5) then the actor is always available
Suitability R1 If
(
PerDa
y
Act < 79.2 and RDAAct < 58.3 and WTAct
(
A
)
< 3
)
then the actor has low skills
R2 If
(
PerDa
y
Act < 79.2 and RDAAct < 58.3 and WTAct
(
A
)
>= 3 and AVDa
y
act < 72.5
)
then the actor has low skills
R3 If
(
PerDa
y
Act >= 79.2 and 37.5<RGAAct < 75 and WTAct
(
A
)
< 12.5
)
then the actor has low skills
Cost R1 If CosDayAct(L) < 4.8 then the cost of the actor is Cheap
R2 If CosDa
y
Act
(
L
)
>= 4.8 and CosDa
y
Act
(
P
)
< 10.16 :then the cost of the actor is Acce
p
table
R3 If CosDayAct(L) < 9 and CosDayAct(P) >= 10.16 then the cost of the actor is Acceptable
Table 7: Identified thresholds values for the evaluation of the characteristics related to BPMN elements.
BPMN
elements
Performance measures Time behaviou
r
Performance
measures
Cost
Threshold Linguistic
inter
p
retation
Threshold Linguistic
inter
p
retation
Activity AD Activity AD<6.5 Low CA
Act
Activity CA
Act
<4.92 Low
6.5<=AD<14.5 Moderate 4.92<=CA
Act
<10 Moderate
AD >=14.5 Hi
g
hCA
Act
>=10 Hi
g
h
Process AD<19.5 Low Process CA
Act
< 18.67 Low
19.5<=AD<28.5 Moderate 8.67 < CA
Act
<= 24.95 Moderate
AD>=28.5 Hi
g
hCA
Act
>=24.95 Hi
g
h
Gateway GD GD <2.5 Low CosGat(Gateway) CosGat < 0.45 Low
2.5<=GD<4.5 Moderate 0.45<=CosGat <0.97 Moderate
GD>=4.5 Hi
g
h CosGat >= 0.97 Hi
g
h
Sequence
Flow
SeqFD Se
q
FD < 4.5 Low CosSeqF CosSe
q
F < 0.45 Low
4.5<= Se
q
FD <8 Moderate 0.45<= CosSe
q
F < 0.99 Moderate
Se
q
FD >= 8 Hi
g
hCosSe
q
F >= 0.99 Hi
g
h
Table 8: Extract of decision rules to assess the level of each characteristic’s BPMN element.
BPMN
elements
Rule Time behaviour Cost
Activit
y
Activity R1 If AD<6.5 then the time of the activit
y
is Minimal If CA
Act
<4.92 then the cost of the activit
y
is Chea
p
R2 If 6.5<=AD<14.5 then the time of the activit
y
is Normal If 4.92<=CA
Act
<10 then the cost of the activit
y
is Acce
p
table
R3 If AD >=14.5 then the time of the activit
y
is Maximal If CA
Act
>=10 then the cost of the activit
y
is Ex
p
ensive
Process R1 If AD<19.5 then the time of the
p
rocess is Minimal If CA
Act
< 18.67 then the cost of the
p
rocess is Chea
p
R2 If 19.5<=AD<28.5 then the time of the process is Normal If 18.67 <CA
Act
<= 24.95 then the cost of the process is Acceptable
R3 If AD>=28.5 then the time of the
p
rocess is Maximal If CA
Act
>=24.95 then the cost of the
p
rocess is Ex
p
ensive
Gateway R1 If GD <2.5 then the time of the
g
atewa
y
is Minimal If GD <2.5 then the cost of the
g
atewa
y
is Minimal
R2 If GD>=2.5 then the time of the
g
atewa
y
is Normal If 2.5<=GD< 4.5 then the cost of the
g
atewa
y
is Normal
R3 If GD>=4.5 then the time of the
g
atewa
y
is Maximal If GD>=4.5 then the cost of the
g
atewa
y
is Maximal
Sequence
Flow
R1 If SeqFD < 4.5 then the time of the sequence flow is
Minimal
If CosSeqF < 0.45 then the cost of the sequence flow is Cheap
R2 If 4.5<= SeqFD <8 then the time of the sequence flow is
Normal
If 0.45<= CosSeqF < 0.99 then the cost of the sequence flow is
Acce
p
table
R3 If SeqFD >= 8 then the time of the sequence flow is
Maximal
If CosSeqF >= 0.99 then the cost of the sequence flow is Expensive
In summary, these thresholds persist imprecise
since they are influenced by the judgment of experts
when we collect the database (Section 3.1.1). To
handle this problem, we use the fuzzy logic.
4 FUZZY LOGIC FOR BP
PERFORMANCE ASSESSMENT
Fuzzy logic is based on the observation that people
make decisions based on imprecise and non-
numerical information. In this paper, we use fuzzy
logic to handle approximate and imprecise values like
those for the performance measures thresholds.
Fuzzy-logic application followed three major steps:
fuzzification, inference and defuzzification.
Fuzzification operations used membership
functions that can map mathematical input values
representing the performance measures into fuzzy
membership functions expressing linguistic values
(i.e. High, moderate, low).
The inference is based on a set of fuzzy decision
rules written in a linguistically natural language.
These fuzzy rules are obtained based on the rules
presented in Section 3.3.
A Methodology for Determination of Performance Measures Thresholds for Business Process
151

The defuzzification produces crisp values of each
performance measure.
In this section, we present in detail how we use the
fuzzy logic to evaluate the performance of BP.
4.1 Fuzzification
Fuzzification transforms crisp values of performance
measures representing the input variables into
linguistic values that express fuzzy sets. This
transformation is realized thanks to the membership
functions that are determined based on the identified
approximate thresholds (Section 3.3). One
membership function is proposed for each possible
fuzzy set per performance measure.
The first part of Figure 2 expresses the
membership without fuzzification. In this part, the
two values (x and y) express the approximate
thresholds obtained based on the use of decision trees
and fuzzy sets that are defined in different intervals.
These intervals are determined by experts (i.e., low,
moderate, high).
Figure 2 illustrates that each performance measure
value, which has a membership degree equals to 1,
belongs only to a single fuzzy set.
Figure 2: Membership function definition.
This situation is true if the identified thresholds
are exact and precise. Nevertheless, since this case
cannot be applied, we use in this paper, the
membership function with fuzzification reflects the
ranges by different experts. It is depicted in the
second part of Figure 2. In this part, the experts
defined four values (x’, x”, y’, y”) for each
performance measure. Each value inside the interval
[x’, x”] and [y’, y”] fits respectively in two fuzzy sets
with different membership degrees. For instance, the
value "z" belongs to the two fuzzy sets "low" and
"moderate" with membership degree of "z1" and
"z2".
4.2 Inference
The inference step used a set of fuzzy decision rules
written in a linguistically natural language. A fuzzy
rule is a simple IF-THEN rule with a condition and a
conclusion. It should be written based on the
following syntax: "if D is X and/or E is Y then F is
Z". D and E represent the input variables, F is the
output variable, and X, Y, Z are their corresponding
linguistic values. 
These rules are crucial to determine the values of
the output variables representing levels of each
BPMN element characteristics (time behavior and
cost), and levels of the actor characteristics (his
availability, suitability and cost) based on the input
values expressing the set of performance measures.
To obtain the fuzzy rules, we start by using the set
of decision rules obtained from the decision tree
(Section 3.3). We changed the crisp values of
performance measures with their corresponding
linguistic values and rewrote the rules according to
the syntax defined above. Table 9 shows the total
number of defined fuzzy rules for each actor’s and
BPMN element characteristic.
Table 9: Total number of defined fuzzy rules for the actor
characteristic and those corresponding to BPMN elements.
Total number of
fuzz
y
rules
Actor Availability 50
Suitabilit
y
207
Cost 15
BPMN
element
Time
behaviour
activit
y
12
Gateway 6
Se
q
uence Flow 6
Cost activity 12
Gatewa
y
6
Se
q
uence Flow 6
Table 10 (respectively Table 11) depicts an
extract of the defined fuzzy decision rules for the
availability, suitability and cost of the actor
(respectively time behaviour and cost of each BPMN
element).
4.3 Defuzzification
Defuzzification is the process of producing a
quantifiable result in crisp logic, given fuzzy sets and
corresponding membership degrees. This conversion
is ensured thanks to a set of membership functions
that we defined based on several rules that transform
a number of variables into a fuzzy result, that is, the
result is described in terms of membership in fuzzy
sets.
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
152

Table 10: Extract of fuzzy decision rules to assess the level of availability, suitability and cost of the actor.
Fuzzy
Rule
Fuzz
y
decision rules
Availabilit
y
Suitabilit
y
Cost
FR1 If (AVDayact is low and PPTAct(A) < 9 and
WTAct(A) < 3) then the AvailabilityLevel is
rarel
y
available
If (PerDayAct is low and RDAAct is low and
WTAct(A) is very low) then the SuitabilityLevel
is havin
g
low skills
If CosDayAct(L) is low then the CostLevel
is Cheap
FR2 If (AVDayact is low and PPTAct(A) >= 9
and PerDayAct >= 78 and TWTDayAct(L) <
15.5) then the AvailabilityLevel is always
available
If (PerDayAct is low and RDAAct is low and
WTAct(A) is low and AVDayact is low ) then
the SuitabilityLevel is having low skills
If CosDayAct(L) is moderate and
CosDayAct(P) is low then the CostLevel is
Acceptable
FR3 If (AVDayact is high and PPTAct(A) < 7 and
PPTAct(A) < 5) then the AvailabilityLevel is
always available
If (PerDayAct is low and RDAAct is low and
WTAct(A) is low and AVDayact is high and
TWTDayAct(L) is low) then the
Suitabilit
y
Level is havin
g
low skills
If CosDayAct(L) is moderate and
CosDayAct(P) is high then the CostLevel
is Acceptable
Table 11: Extract of fuzzy decision rules to assess the level of time behavior and cost of each BPMN element.
BPMN elements Fuzzy
Rule
Time behaviour Cost
activity Activity FR1 If AD is low then the TimeBehaviorLevel is Minimal If CA
Act
is low then the CostLevel is Cheap
FR2 If AD is moderate then the TimeBehaviorLevel is Normal If CA
Act
is moderate then the CostLevel is Acceptable
FR3 If AD is high then the TimeBehaviorLevel is Maximal If CA
Act
is high then the CostLevel is Expensive
Process FR1 If AD is low then the TimeBehaviorLevel is Minimal If CA
Act
is low then the CostLevel is Chea
p
FR2 If AD is moderate then the TimeBehaviorLevel is Normal If CA
Act
is moderate then the CostLevel is Acceptable
FR3 If AD is high then the TimeBehaviorLevel is Maximal If CA
Act
is high then the CostLevel is Expensive
Gateway FR1 If GD is low then the TimeBehaviorLevel is Minimal If CosGat is low then the CostLevel is Chea
p
FR2 If GD is moderate then the TimeBehaviorLevel is Normal If CosGat is moderate then the CostLevel is Acce
p
table
FR3 If GD is hi
g
h then the TimeBehaviorLevel is Maximal If CosGat is hi
g
h then the CostLevel is Ex
p
ensive
sequence flow FR1 If Se
q
FD is low then the TimeBehaviorLevel is Minimal If CosSe
q
F is low then the CostLevel is Chea
p
FR2 If SeqFD is moderate then the TimeBehaviorLevel is
Normal
If CosSeqF is moderate then the CostLevel is Acceptable
FR3 If SeqFD is high then the TimeBehaviorLevel is Maximal If CosSeqF is high then the CostLevel is Expensive
Several defuzzification techniques are proposed
in the literature such as Center Of Gravity (COG),
Centroid Of Area (COA), Mean Of Maximum
(MOM), Center Of Sums (COS), etc. We use the
Center Of Sums (COS) since it is faster than many
defuzzification methods that are presently in use. In
addition, the method is not restricted to symmetric
membership functions. The defuzzified value X
*
of
the output variable is given by equation 5:
∑
∑∑
m
1=j
iA
M
1=i
m
1=j
iAi
*
)X(μ
)X(μ*X
=X
j
j
(5)
Where m is the number of fuzzy sets, M represents
the number of fuzzy variables and expresses the
membership function for the j-th fuzzy sets.
Defuzzification determines the level of each actor
and BPMN element characteristic as well as the
degree of certainty of each level. For example, an
actor can be estimated as the most suitable having
best skills with a certainty degree of 80%.
5 FUZZYPER: FUZZY
PERFORMANCE TOOL
We present in this section our tool that implements
the proposed methodology. The validation is based on
the experimental evaluation.
5.1 FuzzPer Tool
We have developed a tool, bapized "FuzzPer" for
evaluating the cost and time of BPMN elements and
assessing the suitability, availability and cost of the
actor named. Our tool is implemented in Java as an
EclipseTM plug-in (eclipse, 2011). It is composed of
four main modules: Extractor, Measures calculator,
Decision Maker and Fuzzy-logic control as illustrated
in Figure 3.
The extractor takes as input a business process
modeled by BIZAGI tool (ISO/IEC 19510, 2013)
transformed into XPDL file (Shapiro, 2006).
Based on the generated file, the information extracted
by the extractor reflects the semantic (cost and
organizational aspects), temporal and the structural
information. This information involves all BPMN
elements contained in the business process model and
the actors. The use of the standard ensures that our
A Methodology for Determination of Performance Measures Thresholds for Business Process
153

Figure 3: Architecture of FuzzPer tool.
tool can be integrated within any other modeling tool
that supports this standard.
The measures calculator takes as input the XPDL
file, calculates and displays the crisp values of each
used performance measures for estimating either the
cost and time of BPMN elements or the suitability,
availability and cost of the actor.
The Decision Maker takes the crisp values of
performance measures representing the input
variables and transfers them to the fuzzy control
module. This module runs the Fuzzy Control
Language (FCL) for approximating the performance
of BPMN elements and the actor.
Fuzzy-logic Control is implemented in Fuzzy
Control Language (FCL) which is a standard for
Fuzzy Control Programming. It was standardized by
IEC 61131-7. FCL is composed of four main
modules: Function Block Interface, Fuzzification,
Rule identifier, Defuzzification.
 Function Block Interface: determines input and
output parameters.
 Fuzzification: transforms crisp values of
performance measures representing the input
variables into linguistic values (fuzzy sets) that
will be used by the inference engine (Section
4.1). This transformation is realized thanks to
the membership functions that are determined
based on the identified approximate thresholds.
 Rule identifier: is based on a set of fuzzy
decision rules written in a linguistically natural
language to determine the values of the output
variables representing the level each BPMN
element characteristic (time behavior and cost),
and the level of the actor characteristic (his
availability, suitability and cost) (Section 4.2).
 Defuzzification: determines the level of
availability, suitability and cost of the actor and
the level of time behaviour and cost of each
BPMN element as well as the degree of
certainty of this level using the Center Of Sums
(COS) technique (Section 4.3).
Based on the obtained result provided by the
Defuzziification, the decision maker estimates the
performance of the actor and BPMN elements.
5.2 Experiments
In order to validate our methodology, we are based on
two experiments. The first experiment involved
students from our college while the second
experiment use "FuzzPer" tool. These experiments
use the following additional resources:
 Business Process Model: we use the "Travel
Agency process" example modelled with
BPMN in Figure 4. The model is annotated by
temporal constraints and semantic information
(cost and organizational aspects).
 Participants: During these experiments, we
asked 50 students from our faculty to assess the
actor characteristics (availability, cost
suitability) and to evaluate BPMN elements
characteristics (time behaviour and cost).
 Actor characteristics exercise: students should
respond to a set of questions to evaluate the
performance of the actor. The questions are
classified into three categories: those that focus
on the availability of the actor, those related to
the suitability of the actor and those related to
the actor cost. Finally, each student should
select the level of each actor characteristic (i.e.
availability, suitability and cost). For instance,
the actor is always available or rarely available.
In addition, he has the best skills or low skills.
The exercise is available at: https://sites.
google.com/site/kchaoumariemsint/resources.
 BPMN elements characteristics exercise:
students had to evaluate the time of performing
an activity, the time of make decision and the
transfer time. Finally, each one choose the time
level and the cost level of each BPMN element.
For instance, the activity’s cost is cheap,
acceptable or expensive. The exercise is
available at: https://sites.google.com/site/
kchaoumariemsint/resources
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
154

Figure 4: "Travel Agency process".
Experiment 1: According to the actor character-istics
exercise, 75% of the responses are correct. The result
expresses that the majority of students can assess the
performance of the actor. This result is also established
based on their responses to the last question for each
category of the first exercise, which is about the
availability, suitability and cost of the actor.
Indeed, 78% of students considered the actor as
always available, 22% as rarely available. In addition,
69% of the responses are correct. They show that a
good number of students have correctly evaluate
competences of actors having low skills. In addition,
52% of students considered the actors as expensive,
29 % as acceptable, and 19% as cheap.
According to BPMN elements characteristics
exercise, 67% of the responses are correct. Based on
this result, we can deduce that the majority of students
can evaluate the performance of BPMN elements.
This result is also established based on their responses
to the last question for each category of the second
exercise, which is about the time behaviour and cost
of BPMN elements.
Indeed, 61% of students consider the time of
activities in the BPMN model as normal, 23% as
maximal and 16 % as minimal. In addition, 74% of
students considered activities as expensive, 16 % as
acceptable, and 10% as cheap.
Experiment 2: Uses our tool to estimate the actor
characteristics levels and the BPMN element
characteristics levels of the business process model
illustrated in Figure 4. Our BPMN model is annotated
by temporal constraints and semantic information
(cost and organizational aspects).
Considering the limited space, we present an
example of the actor characteristic (suitability) and an
example of BPMN element characteristic such as the
time behaviour of an activity.
For instance, the estimated suitability level of the
actor “Omar” is “Having Low Skills with a certainty
degree of 63%”. Figure 5 shows the interface for
actor suitability assessment.
Figure 5: Availability characteristic assessment interface.
In addition, the estimated time behaviour level of
the BPMN element “Activity” is “Normal with a
certainty degree of 67%”. Figure 6 shows the
interface for the activity assessment.
Actor Ali
AD 8 mi
n
PPT 9 mi
n
WT 8 mi
n
State Defecte
d
CA 2.4 euro
GD 2 mi
n
CosGat 0.7 euro
Actor Relationship Labour Cost
Khadija Khadija is the leader of Ali 21 euro
Ali Ali is under the hierarchy of Khadija 18 euro
Omar Omar has the same position of Khadija 21 euro
Actor Omar
AD 11 mi
n
PPT 13 mi
n
WT 11 mi
n
State Goo
d
CA 3.85 euro
Actor Omar
AD 7 mi
n
PPT 11 mi
n
WT 7 mi
n
State Defecte
d
CA 2.45 euro
Actor Omar
AD 9 mi
n
PPT 13 mi
n
WT 9 mi
n
State Goo
d
CA 3.15 euro
Actor Omar
AD 3 mi
n
PPT 5 mi
n
WT 3 mi
n
State Goo
d
CA 1.05 euro
Actor Omar
AD 7 mi
n
PPT 10 mi
n
WT 7 mi
n
State Goo
d
CA 2.45 euro
GD 1 min
CosGat 0.35 euro
AD Activity Duratio
n
PPT Panned Production Time
WT Working Time
State Good or Defecte
d
CA Cost of an Activit
y
GD Gateway Duaratio
n
CosGat Cost of a Gatewa
y
Actor Khadija
AD 7 mi
n
PPT 11 mi
n
WT 7 mi
n
State Defecte
d
CA 2.45 euro
Actor Khadija
AD 13 mi
n
PPT 18 mi
n
WT 13 mi
n
State Goo
d
CA 4.55 euro
A Methodology for Determination of Performance Measures Thresholds for Business Process
155

Figure 6: Time behaviour characteristic assessment
interface.
Based on responses of students to the suitability
questions, experiments reveal that the suitability of
actors as having low skills. This result is conform
with the evaluation effected by FuzzPer, which
reflects that the actors in the presented BP model as
“having low skills”.
As the same, regarding the time
behaviour, students consider the time of activities in
the BPMN model “Travel agency process” as normal.
These compliant results demonstrate that our
methodology provides promising results that should
be shown based on further experiments.
6 THREATS TO VALIDITY
This study, as every other empirical business process
study, is subject to two type of threats: internal,
external (Wohlin et al., 2000).
The internal validity threats are related to the
following issues: The first issue is the use of three
algorithms (J48, RandomTree, REPTree) to find the
imprecise thresholds using our methodology. We
chose REPTree algorithm for finding threshold
values as it is the one that yielded the best results. Of
course, we should find other algorithms to determine
more objectively the values of thresholds. The second
issue is that although the annotation of BPMN models
are listed in the datasets used, these information has
not been tested. Therefore, some errors may not have
been discovered in some BPMN models. Considering
this, our thresholds could have found faults that are
yet undiscovered.
The external validity is related to the limited
number of the used databases (one datbase). Our
study covers only BPMN models having small/
medium size. This means that the findings of this
study cannot be generalized to all BPMN models,
particularly those having complex size. Further tests
on many other BP from different domains would be
needed to generalize obtained results.
7 CONCLUSION
In this paper, we proposed a methodology to assess
the performance of actors and BPMN elements. Our
methodology use a set of performance measures. It
followed two major phases: threshold determination
and fuzzy logic application.
The first phase “Threshold determination” is
based on “Business process database”. It is composed
of two stages: Analyze data and Validate Data. The
first stage starts by collecting a set of BPMN models
annotated by semantic and temporal information, then
preparing these models through creating matrices
related to actors and to BPMN elements to evaluate
their characteristics and finally, applying Data
mining to build decision trees using WEKA system.
The system determine approximate thresholds for
each used performance measures.
The second phase of the proposed methodology
uses fuzzy logic to handle approximate and imprecise
nature of the obtained thresholds in the first phase. To
automate BP models performance evaluation, we
developed a FuzzPer tool. To illustrate the efficiency
of the performance tool, we rely on two types of
experimental evaluation. The former is accomplished
with students while the second is done through the
proposed tool.
The preliminary experiments’ results of the
proposed tool display encouraging results related to
the evaluation of the actor and BPMN elements
performance.
Our future work focuses on two main axes: 1)
validate the proposed fuzzy methodology for BP
performance evaluation through some real case
studies with business experts and 2) evaluate the
performance of actors and BPMN elements in terms
of other characteristics.
REFERENCES
Chen, J., Wang, X., Zhai, J., 2009. Pruning Decision Tree
Using Genetic Algorithms. Inter. Conf. on Artificial
Intelligence and Computational Intelligence, pp. 244-
248.
D'Ambrogio, A., Paglia, E., Bocciarelli, P., Giglio,A.,
2016. Towards performance-oriented perfective
evolution of BPMN models. In SpringSim'16, 49th
Inter.Conf. on Spring Simulation Multi-Conf, pp. 15.
ENASE 2020 - 15th International Conference on Evaluation of Novel Approaches to Software Engineering
156

Eclipse Specification, 2011. Available at: www.eclipse.org/
documentation.
Del-Río-Ortega, A., Resinas, M., Cabanillas, C., Ruiz
Cortés, A. 2013. Defining and Analysing Resource-
Aware Process Performance Indicators. In CAiSE’25,
International Conference on Advanced Information
Systems Engineering, Ljubljana, Slovenia, pp. 57-64.
Gonzalez-Lopez, F., Bustos, G., 2019. Business process
architecture design methodologies - a literature review.
In International Journal of Business Process
Management, pp. 1317-1334.
Hall, M., Frank, E., Holmes, G., Pfahringer, B., Reutemann,
P., and Witten, I. H., 2009. The weka data mining
software: an update. ACM SIGKDD explorations
newsletter, pp.10–18.
Heinrich, R., Paech, B., 2010. Defining the Quality of
Business Processes. In Modellierung’10, 4
th
Inter.
Conference of Modellierung, pp.133-148.
Heinrich, R. 2013. Aligning Business Process Quality and
Information System Quality. PhD thesis.
ISO/IEC 19510, 2013 - information technology - object
management group business process model and
notation.
Kchaou, M., Khlif, W., Gargouri, F., 2019. Temporal,
Semantic and Structural Aspects-based Transformation
Rules for Refactoring BPMN Model. In the 16
th
Inter.
Joint Conference on e-Business and
Telecommunications, Prague, Czech Republic. pp. 133-
144.
Khlif, W., Kchaou, M., Gargouri, F., 2019. A Framework
for Evaluating Business Process Performance. In the
14
th
Inter. Conference on Software Technologies,
ICSOFT 2019, Prague, Czech Republic, pp. 371-383.
Kis, I., Bachhofner, S., Di Ciccio, C., Mendling, J., 2017.
Towards a Data-Driven Framework for Measuring
Process Performance. In BPMDS’17, 18
th
Inter.
Conference on Enterprise, Business-Process and
Information Systems Modeling, Germany, pp. 3-18
Kluza, K., Nalepa, J.G., 2019. Formal Model of Business
Processes Integrated with Business Rules. In Inter.
Jour. of Information Systems Frontiers, pp1167–1185.
Lanz, A., Reichert, M., Weber, B., 2016. Process time
patterns: A formal foundation. In International Journal
of the Information Systems, V. 57, pp. 38-68.
Razzaq, S., Shujahat, M., Hussain, S., Nawaz, F., Wang,
M., Ali, M., Tehseen, S., 2019. Knowledge
management, organizational commitment and
knowledge-worker performance. In Inter. Journal of
Business Process Management, V. 25, pp.923-947.
Shapiro, R.M., 2006. XPDL 2.0: Integrating process
interchange and BPMN. pp. 183–194
Van der Aa, H., Del-Río-Ortega, A., Resinas, M., Leopold,
H., Cortés,A.R., Mendling, J., Reijers, H.A., 2016.
Narrowing the Business-IT Gap in Process
Performance Measurement. In CAiSE’16, 28
th
Inter.
Conference Advanced Information Systems
Engineering, Ljubljana, Slovenia, pp. 543-557.
Wohlin, C., Runeson, P., Höst, M., Ohlsson, M. C.,
Regnell, B., Wesslén, A., 2000. Experimentation in
Software Engineering: An Introduction. Kluwer,
Academic Publishers.
Wynn, M. T., Low, W. Z., Nauta, W., 2013. A Framework
for Cost-Aware Process Management: Generation of
Accurate and Timely Management Accounting Cost
Reports. In APCCM’9, Asia-Pacific Conference on
Conceptual Modelling, pp. 79-88.
Zadeh, L. A., 1965. Fuzzy sets. Information and control. In
Inter.l Jour. of Infor. and Control, pp.338–353.
Zadeh, L. A., 2008. Is there a need for fuzzy logic? In Inter.
Jour. of Information sciences, pp. 2751–2779.
A Methodology for Determination of Performance Measures Thresholds for Business Process
157