Guided Process Discovery Approach According to Business Process

Types

Lamghari Zineb

, Saidi Rajaa

1,2

, Radgui Maryam

1,2

and Rahmani Moulay Driss

LRIT associated unit to CNRST (URAC 29) Rabat IT Center Faculty of Sciences, Mohammed V University in Rabat,

Morocco

SI2M Laboratory, National Institute of Statistics and Applied Economics, Morocco

Keywords:

Process mining, process discovery, business process type, process mining techniques, process model perspec-

tives.

Abstract:

Process discovery technique aims at automatically generating a process model that accurately describes a

Business Process (BP) based on event data. Related discovery algorithms consider recorded events are only

resulting from an operational BP type. While the management community deﬁnes three business process

types, which are: Management, Support and Operational. They distinguish each BP type by different propri-

eties like the main business process objective as domain knowledge. This puts forward the lack of process

discovery technique in obtaining process models according to business process types (Management and Sup-

port).

In this paper, we demonstrate that business process types can guide the process discovery technique in gener-

ating process models. A special interest is given to the use of process mining to deal with this challenge.

1 INTRODUCTION

Business Processes (BPs) are nowadays a crucial el-

ement in any organizational structure. They are es-

tablished to manage and improve the company busi-

ness. In this context, information systems assure the

automation of BPs (Lamghari et al., 2021), by includ-

ing Business Process Management (BPM) systems,

Enterprise Resource Planning (ERP), Customer Re-

lationship Management (CRM), etc. In this sense, in-

formation systems record related event data to the BP

execution, to analyse and guide issues concerning the

creation of the company value. To achieve these ob-

jectives, process mining techniques are created.

Process mining is a scientiﬁc discipline that fo-

cuses on the analysis of event data logged during the

execution of a BP, in order to discover, monitor and

enhance BPs. It (Eck et al., 2015) consists of two

major phases: Pre-processing and Processing. De-

pending on the process subject, we extract data from

Repository. Once the data has been extracted; we pro-

ceed to ﬁlter it from deﬁciencies in the log cleaning

step. By doing so, we arrive to the process model dis-

covery technique. Next, we apply the conformance

checking step, which aims at evaluating how well this

discovered process model corresponds to reality. This

gives input for changing parameters in the main pro-

cess, and requires going back to the data cleaning

step, to re-apply the discovery technique. After eval-

uation, we diagnose all obtained results, to provide

inputs for business process improvements. Therefore,

event data analysis and the discovery technique are

required for obtaining suitable process models that

takes into consideration: event logs content, their lev-

els of abstraction (Aalst, 2016) and the process model

perspectives (Bogarin et al., 2018). By necessity, a

suitable process model reﬂects the business reality

and met requirements.

Process model perspectives are: Control-ﬂow

(The order in which its activities should be executed),

Organizational (The resources required for the exe-

cution of a process and how they interact with each

other), data-ﬂow (The data objects created and up-

dated during the execution of the process) and time

(The time-related aspects of the process). In this

work, we assume that the pre-processing phase works

directly on cleaning event logs, while the processing

phase focuses on analysing events and constructing

process models.

In this sense, depending on the input data and the

questions that need to be answered, a suitable pro-

cess model abstraction can be presented. The model

may be too abstract and thus unable to answer relevant

questions. The model may also be too detailed, e.g.,

the required input cannot be obtained or the model

becomes too complex to be fully understood.In this

context, many process discovery methods assume that

recorded events correspond to meaningful activities

in the instances of a process. However, events may

be recorded on different levels of granularity. Some

events may refer to activities on a high level of ab-

straction. Their execution is easily recognizable for

process workers. Other events may be recorded with

a lower level of abstraction. Multiple of such low-

level events may referee to a recognizable high-level

activity. When discovering processes based on those

low-level events, the resulting process model can im-

pact process workers structure. Consequently, the dis-

covered model represents the wrong level of abstrac-

tion. Therefore, the event data Domain Knowledge

(DK) is the crucial parameter that can guide the pro-

cess discovery technique Ê, in terms of diagnosing

and representing process models according to their

DK. Besides, the management community has de-

ﬁned the BP type as the DK that can impact (Harmon,

2015; Lamghari et al., 2018) the BP representation

and treatment Ë.

Based on these two information Ê and Ë, we assume

the following hypothesis: BP type as DK may inﬂu-

ence process model that can be discovered with pro-

cess discovery techniques, in terms of perspectives

priority, i.e., according to DK which perspective will

be treated or combined ﬁrstly with the control-ﬂow

perspective for guiding the process discovery in gen-

erating process models.

The BP types (Burattin, 2015) are:

• The Management BP: Describes the process of the

product or services realization that is provided by

the company to their customers.

• The Operational BP: Deﬁnes the organization

strategy.

• The Support BP: presents the process that offers

resources to other processes to ensure the smooth

running of the company.

Thus, we need an approach that can tackle the

treatment of events and discover models according to

BP types (Burattin, 2015), towards guiding the pro-

cess discovery technique in discovering suitable mod-

els. This can be achieved using DK as the BP type

related to multiple process model perspectives.

In this context, we must discuss different related

issues that involve the intersection between BP types

and process mining. For this purpose, our paper pro-

poses an approach that demonstrates how BP type, as

DK, will guide and impact the process discovery pro-

cess. In this sense, our paper is organized as follows:

Section 2 presents the still encountered related issues

of the intersection between BP types and process min-

ing. The section 3 details our approach that consists

of guiding the process discovery technique according

to BP types. Conclusion and further directions are

presented in section 4.

2 BRIDGING THE GAP

BETWEEN BUSINESS

PROCESS TYPES AND

PROCESS MINING

In this section, we present the still encountered issues

related to the intersection between BP types and pro-

cess mining, which are: Event data quality, Manage-

ment view (Conﬁguration phase, BP status, and BP

execution), Levels of representing process models or

their abstraction that includes process model perspec-

tives and the clustering technique.

2.1 Event Data Quality

The quality is achieved by ﬁltering out noise (incor-

rectly logged), incomplete (missing events), chaotic

(arbitrary executed) and infrequent behaviours from

event data. These deﬁciencies are denoted, in

Fig.1, respectively as (Domain

knowledge N, (Do-

main knowledge I, etc).

In the literature, we ﬁnd many research papers,

like (Conforti et al., 2017; s. Suriadi et al., 2017),

dealing with this topic. This issue has received a sig-

niﬁcant focus from the research community, where all

related ambiguities have been resolved. Here, we pro-

pose to use the BP type as an additional ﬁltering pa-

rameter, to reﬁne the main BP context.

2.2 Management View

Information systems record event data. Their conﬁg-

uration is one of the most prominent tasks related to

event data preparation.

Generally, a BP is deﬁned with speciﬁc parameters as

DK (BP type), BP status (informs about the BP life-

cycle) and other proprieties that depend on the BP

objective and the organization requirements. These

parameters describe the conﬁguration ﬁle content, to

record behaviours and avoid errors that may be pro-

vided during the BP execution. The point has not

matched, in this regard, is the deﬁnition of the BP

type, its particularity and how it may impact the pro-

cess discovery technique and the process model per-

Figure 1: Related issues to the intersection between BP types and process mining.

Figure 2: Our contribution aspects.

spectives, in the term of representation. Available re-

search (Bogarin et al., 2018; Conforti et al., 2017;

Boubaker et al., 2016) consider recorded events that

are resulting from an operational BP type. These stud-

ies do not mention the two other BP types (Support

and Management), while the BP type can inﬂuence

the representation of process models that can be dis-

covered with process discovery techniques. In this

respect, the DK or BP type is implicitly deﬁned in

the conﬁguration ﬁle, it has not been declared as at-

tribute in the event object (a part of the conﬁguration

ﬁle). This provides another untreated point to con-

sider. Thus, the propriety we propose to use is to de-

clare BP type as an attribute that will determine the

BP domain knowledge.

2.3 Levels of Abstraction

The process discovery technique generates models us-

ing the 2-D method (Aalst, 2016), where the process

model can be viewed in different levels of abstrac-

tion. As mentioned in the (Bogarin et al., 2018) events

of low-level are mapped to activity instances of high-

level, based on speciﬁc DK. The correspondence be-

tween these two levels is treated and provided ade-

quate results (Bogarin et al., 2018; Baier et al., 2016;

Ciccio et al., 2018). For instance, one high-level ac-

tivity instance may result in multiple low-level events

being recorded and, vice versa, one such low-level

event may relate to multiple high-level activity in-

stances.

The point not yet treated is the BP type (Do-

main knowledge L in Fig.1) impact on the process

model abstraction, by necessity process model per-

spectives’ priority, i.e.; according to the BP type as

DK, which process model perspective will be com-

bined with the control-ﬂow perspective.

2.4 Clustering Technique

The clustering technique is treated in the processing

phase as mentioned in (Oliveira and Queiroz, 2020).

We do not observe it in the pre-processing phase. In-

troducing this technique into the pre-processing phase

can reﬁne the ﬁnal process model construction. More-

over, this technique based on DK as BP type deﬁni-

tion that groups a set of activity instances with the

same context (BP type), to determine a clustered pro-

cess model. In this sense, for event logs sequences,

we can determine the set of similar data (cluster).

2.5 Synthesis

According to this discussion, we have detected some

untreated stages related to BP types (Cf. Fig.1).

These questionable stages emerged new issues, rel-

atively to process mining. We have also explored

an additional treated phase, which is the conﬁgura-

tion phase. Therefore, our contribution will focus on

three phases (conﬁguration, pre-processing and pro-

cessing). In this sense, DK is our guided approach

parameter. Thus, we will deﬁne DK as (Cf. Fig.2 and

Fig.1):

• BP type presents event object in the conﬁguration

phase (Domain knowledge G).

• BP type presents an additional ﬁlter to deﬁne

the main cluster for pre-processing phase (Do-

main knowledge C).

• BP type as BP type objective and related to pro-

cess model perspectives (Domain knowledge G).

3 OUR GUIDED PROCESS

DISCOVERY APPROACH

Our approach aims at guiding the process discovery

technique in representing process models according

to different BP types. It consists of three phases: con-

ﬁguration, pre-processing and processing (Cf. Fig.3).

Our approach process starts by conﬁguring the

information system, where all BPs will be executed

and recorded according to the event object parame-

ters. Once, the selected BP is executed, we verify if

the conﬁguration elements are successfully achieved.

If this later is well done, we proceed to the pre-

processing phase, which aims at cleaning event data

from deﬁciencies. If not, we loop back to the conﬁgu-

ration phase. Finally, we pass to the processing phase,

to obtain process model representation.

Throughout our approach phases:

• We add a new event object in the conﬁguration

phase À. After executing a BP Á and collecting

recorded event logs, we arrive to the second oper-

ation named Â that uses an additional ﬁltering at-

tribute, which is the BP type of the pre-processing

phase. Finally, we apply Ã the processing phase

techniques for mining process models according

to BP types, by necessity the BP type objective

and the process model perspectives (control-ﬂow,

time, organizational).

• We deﬁne different DK, and we obtain different

outputs (1-main frame based on the BP type, 2-

main cluster based on the BP type, 3-clustered

process model based on the BP type).

3.1 Conﬁguration Phase

The conﬁguration phase (Cf. Fig.3-À) takes as an in-

put a BP model. It gives idea and insights about the

BP type. Indeed, the conﬁguration phase takes into

consideration the BP type as DK.

The reason to choose the BP type as DK within the

conﬁguration phase is:

• The use of management view from early stage.

Figure 3: Guided Process Discovery approach phases.

• The BP type impacts on the next two phases.

• Preparing events in the conﬁguration phase as an

early basic ﬁlter.

The conﬁguration process consists of:

• Determining the BP domain knowledge as the op-

erational BP, the management BP, and the support

BP.

• Deﬁne the event object content.

• Declare the BP type as attribute in the event ob-

ject.

• Activate the recording option.

3.2 Execution Phase

During the execution phase (Cf. Fig.3-Á), we record

BP event logs. In this context, the minimum infor-

mation required of a standard format for an event log

is: The case ID is a unique identiﬁer for a process

instance, the information stated in chronological or-

der according to events. Also, additional information

is possible, we ﬁnd attributes such as timestamp (the

time when the activity took place), resources (who

performed the activity), transaction type and costs as-

sociated with the event.

3.3 Pre-processing Phase

After executing the BP in (Cf. Fig.3-Â), the pre-

processing process (Cf. Fig.3-Â) starts by extracting

events in the adequate form, to apply the ﬁltering op-

eration. This operation cleans events from deﬁcien-

cies: noise, incompleteness, chaotic and infrequent

behaviours. After obtaining a cleaned event logs, we

deﬁne the main cluster using the BP type parameter

(the main frame of event data). The reason for intro-

ducing a main cluster, in the pre-processing phase, is

to provide an advanced recognition technique using

the BP type.

3.4 Processing

After declaring the BP type attribute in the conﬁgura-

tion phase and deﬁning the main cluster based on the

BP type in the pre-processing phase, we arrive to the

processing phase (Cf. Fig.3-Ã). Therefore, the pro-

cessing phase consists of:

• Using the clustering technique, to get a clustered

process model, i.e., clustered process model based

on the BP type. We deﬁne clusters semantically

(Staab and Studer, 2010).

• Deﬁning the process model perspectives priority:

The BP type gives idea on the main BP objective

Figure 4: Example of process model perspectives order.

and gives insights on the process model perspec-

tives order (In which order perspectives will be

combined with the control-ﬂow) within the logic

of:

The BP type + the BP objective Õ Results the or-

der of representing process model details:

– Operational BP + Satisfaction Õ (1. Time, 2.

Resource, 3. Data) Satisfy the client means that

a process respects time limitation. If not, I will

verify why resource did not respect the time

condition in order to the data-ﬂow into this BP.

– Management BP + Management method Õ (1.

Data, 2. Resource, 3. Time) Achieve the

management method objective in respect to the

data-ﬂow used, into this BP, by resources in

time condition.

– Support BP + Resource Support Õ (1. Re-

source, 2. Time, 3. Data-ﬂow) Obtain re-

sources support related to the time indication

and the data-ﬂow of this BP.

The perspectives order can be veriﬁed in the conﬁgu-

ration phase, and it can provide more insights on the

probably detected deviations.

3.5 Illustrative Example

In this sub-section, we treat an operational BP about

claim’s handler http://www.processmining.org/

event_logs_and_models_used_in_book. We aim

to verify our process model perspectives order propo-

sition.

In our example, the claim’s handler BP makes sure

that claims are handled efﬁciently and that payment

for valid claims is made. Also, this process consists

of making decisions on the extent and validity of a

claim, and the checking for any potentially fraudulent

activity (Cf. Fig.4).

To represent our process model example, we use the

heuristic miner algorithm (Weijters et al., 2006). By

necessity, this algorithm takes into consideration pro-

cess quality metrics: ﬁtness, precision, generaliza-

tion, and simplicity.

According to this BP type, we can ﬁrstly detect the

time deviation. Then, by deﬁning on which activity

the deviation is provided, we can deﬁne responsible

resources for this deviation and respectively data-ﬂow

contributed to this deviation. Therefore, we observe

different advantages of our proposed approach, com-

paring to the ordinal process mining project phases

(Eck et al., 2015):

• BP type as domain knowledge used within three

phases.

• Clustering technique applied from the pre-

processing phase.

• Filtering insights mentioned from the conﬁgura-

tion phase.

• Conformance checking indications appeared from

the discovery step.

• Basic ﬁltering applied from the conﬁguration

phase (By BP types).

• Combining the management view with the pro-

cess model perspectives, which are resulting the

process model perspectives priority concept.

• New process model representation according to

BP types.

3.6 Synthesis

Our approach is applied on three phases: conﬁgu-

ration, pre-processing and processing. Throughout

these phases, we have presented a guided process dis-

covery approach by BP types. The general idea con-

sists of taking into consideration the impact of BP

types on process model representation.

We have treated the still encountered related issues

of the intersection between BP types and process

mining: Management view, Clustering technique and

Process model. The management view takes into con-

sideration the impact of the BP type on event logs

recognition. The clustering technique uses BP type as

context, to group a set of commune activity instances.

The process model perspectives propose a ﬂow of rep-

resentation guided by BP types. In this sense, the con-

formance checking technique can approve the appli-

cation order of process model perspectives according

to BP types.

4 CONCLUSIONS

In this paper, we present an approach dealing with the

process discovery technique according to BP types.

Indeed, we aim to guide the process mining discov-

ery technique, in order to generate suitable process

model for each BP type. For this purpose, we inves-

tigate the still encountered issues related to the inter-

section between BP types and process mining. We ob-

serve four mains required objectives: the management

view, process model perspectives and the clustering

technique. In this context, we match each challenge

with a speciﬁc phase of our proposed approach. Con-

sequently, our approach is applied on three phases:

conﬁguration, pre-processing and processing. In this

respect, the conﬁguration phase declares BP type as

event object, to deﬁne selected event data (Lamghari

et al., 2019) by BP type. Then, the pre-processing

phase use a new ﬁlter, which aims to reﬁne event

data frame by BP type. Last, the processing phase

treats event logs, using the correspondence between

BP types and process model perspectives’ priority,

to represent process models according to BP types.

This helps in acquiring insights on which order per-

spectives could be combined to the control-ﬂow per-

spective. As further research, we plan to develop a

full plug-in that can be implemented into the Prom

tool, for applying our proposed guided process dis-

covery approach according to BP types and improve-

ment metrics (Lamghari et al., 2019).

ACKNOWLEDGEMENTS

This work was supported by the National Center for

Scientiﬁc and Technical Research (CNRST) in Rabat,

Morocco.

REFERENCES

Aalst, W. M. P. V. D. (2016). Data science in action.

Springer, 2nd edition.

Baier, T., Ciccio, C. D., Mendling, J., and Weske, M.

(2016). Matching events and activities by integrating

behavioral aspects and label analysis. International

Journal of Business Process Integration and Manage-

ment, 17(2):573–598.

Bogarin, A., Cerezo, R., and Romero, C. (2018). A sur-

vey on educational process mining. Wiley Interdisci-

plinary Reviews: Data Mining and Knowledge Dis-

covery, 8(1):12–30.

Boubaker, S., Mammar, A., Graiet, M., and Gaaboul, W.

(2016). A formal guidance approach for correct pro-

cess conﬁguration. In International Conference on

Service-Oriented Computing. Springer.

Burattin, A. (2015). Process Mining Techniques in Business

Environments. Springer.

Ciccio, C. D., Maggi, J., Montali, M., and Mendling (2018).

Matching events and activities by integrating behav-

ioral aspects and label analysis. Information Systems,

78(2):144–161.

Conforti, R., Rosa, M. L., and Hofstede, H. T. (2017). Fil-

tering out infrequent behavior from business process

event logs. IEEE Transactions on Knowledge and

Data Engineering, 19(2):300–314.

Eck, M. V., Lu, X., Leemans, S., and aalst, W. D. (2015).

Pm2: A process mining project methodology. In The

27th International Conference on Advanced Informa-

tion Systems Engineering Proceedings. Springer.

Harmon, P. (2015). Business process change: a business

process management guide for managers and process

professionals. Morgan Kaufmann.

Lamghari, Z., Radgui, M., Saidi, R., and Rahmani, M.

(2018). A set of indicators for bpm life cycle improve-

ment. In International Conference on Intelligent Sys-

tems and Computer Vision. IEEE.

Lamghari, Z., Radgui, M., Saidi, R., and Rahmani, M.

(2019). Deﬁning business process improvement met-

rics based on bpm life cycle and process mining tech-

niques. International Journal of Business Process In-

tegration and Management, 9(2):107–133.

Lamghari, Z., Saidi, R., Radgui, M., and Rahmani, M.

(2021). An operational support approach for min-

ing unstructured business processes. Revista de In-

form

atica Te

orica e Aplicada, 28(1):23–38.

Oliveira, M. D. S. and Queiroz, S. (2020). Unsuper-

vised feature selection methodology for clustering in

high dimensionality datasets. Revista de Inform

atica

orica e Aplicada, 27(2):30–41.

s. Suriadi, Andrews, R., and Hofstede, H. T. (2017). Fil-

tering out infrequent behavior from business process

event logs. Event log imperfection patterns for pro-

cess mining: Towards a systematic approach to clean-

ing event logs, 64(2):132–150.

Staab, S. and Studer, R. (2010). Handbook on ontologies.

pringer Science and Business Media.

Weijters, A., Aalst, W. D., and Medeiros, A. D. (2006).

Process mining with the heuristics miner-algorithm.

Technische Universiteit Eindhoven.