Hind R’bigui The Graduate School of the University of Ulsan

마이닝 프로세스의 유형은 파생, 적합성 및 업그레이드입니다. 프로세스 모델 도출은 가장 중요한 프로세스 마이닝 기술입니다.

INTRODUCTION

Process Aware Systems Limitations

Process Mining

In the process discovery category, 3 types of perspectives can be performed based on the information available in the event log. The timestamps shown in Table 1.1 represent the time when the execution of the corresponding activity was started.

Process Mining Challenges

The representation bias used in process discovery greatly affects the quality of process mining results. The purpose of process mining is to produce process models that can be used for further analysis rather than documentation (Buijs, 2014a).

The state of the art of process mining challenges

Finding, Merging, and Cleaning Event Data
Dealing with complex event logs with diverse characteristics
Creating representative benchmarks
Addressing Concept Drift
Improving the Representational Bias Used for Process Discovery
Balancing between quality criteria of fitness, simplicity, precision, and
Cross-Organizational mining
Providing operational support
Combining process mining with other types of analysis
Improving usability for non-experts
Improving understandability for non-experts

Mokhov and Carmona (2014) described the first attempt to use parameterized graphs (PG) in process mining. Buijs et al. (2012a) proposed a novel process mining genetic algorithm that discovers process trees from event logs.

Figure 1.4. Publications dealing with the main process mining challenges

Process discovery techniques limitations overview

The region-based algorithm can extract non-free-choice constructs, but cannot handle invisible tasks and duplicate tasks (Bergenthum et al., 2007). The genetic mining algorithm is the only existing algorithm that can handle most of the common constructs and noisy logs (Van der Aalst et al., 2005).

Contributions and Structure of theThesis

The second major part of the current study, which deals with developing a new process discovery approach capable of dealing with the standard and complex constructions of a process model, is presented in Chapter 6. This process discovery technique is evaluated using both artificial as real life data. in chapter 7.

INDUSTRIAL APPLICATION OF PROCESS MINING

Literature review on process mining case studies

The main focus of this case study was on the conformance verification type of process mining. In (Goedertier et al., 2011) a process mining case study is described about a customer invoice handling process of a telecommunications industry.

Table 2.1. An overview of industrial applications of process mining in the practical world

Case study

Case Description
Application Methodology
Event log construction
Event log filtering
Customer Order Fulfillment Process Discovery
Compliance verification
Process performance analysis

The process model discovered in the previous section can be a guide to improve the standard process model. With the time attribute of events, it is possible to identify bottlenecks that negatively affect the performance of the entire process. Similar to the process of fulfilling customer orders, we extracted data related to the raw material procurement process from the company's SPPM system.

Therefore, the first step to improve the performance of the customer order fulfillment process and reduce its lead time is to reduce the identified waiting time between the raw material purchase request and the production process. This can be done by investigating the root causes of the delays that occurred in the material procurement process.

Figure 2.1. Customer order fulfillment standard process description

PRELIMINARIES

Event logs
Process models

Petri net
Workflow net

Standard constructs of a Petri net
Complex constructs of a Petri net

Invisible Tasks
Short Loops
Duplicate Tasks (DT)
Non-Free Choice Construct (NFC)

In the marker shown in Figure 3.1 (i.e., one token at the source place), the Purchase request transition is enabled, and activating this transition removes the token from the input place and places a token at the output place. We will define these complex constructs in this section before presenting our discovery approach in the next section. An invisible task is a hidden task that has been executed but does not appear in the event log.

Duplicate tasks are tasks that share the same name but are located on two or more nodes in the workflow network. Since it is difficult to distinguish between duplicate tasks that share the same name in the event log, it is difficult to mine such a process model correctly based on its event log.

Figure 3.1. The corresponding petri net model of the event log of table 3.1(A sound workflow net)

PROCESS DISCOVERY TECHNIQUES RECOMMENDATION

Literature review

Based on features extracted from event logs and prediction models built from experiments, top-K control flow miners are recommended. The proposed systems would have been more interesting if the authors had based their framework on event logs and reduced the experimentation required to decide which mining algorithm is appropriate for a given event log. Pérez-Alfonso et al, (2015) proposed an approach to recommend a process discovery algorithm based only on the classification of event logs, but they just conceptualized the idea. 2018) proposed a classification-based framework to evaluate the quality of process discovery algorithms.

The quality of process detection algorithms is based on the algorithm's ability to correctly classify traces that represent actual process behavior as matches and traces that represent behavior unrelated to the process as mismatches. The user should run experiments on all existing detection techniques until he decides on the best algorithm based on quality performance.

Process Discovery Recommendation Framework

Constructs
Complex construct detection
Knowledge data base construction

Finally, relation →𝐿3 detects the implicit dependencies similar to Figure 4.4(a). e) Invisible tasks involved in a non-free choice construct detection. The reachable dependencies associated with invisible tasks involved in a non-free choice construct are defined in Definition 4.18. LS𝑝𝑎𝑟, IvT - SW, non-free choice constructs, and invisible tasks involved in a non-free choice construct.

Finally, the genetic miner (GM) can discover all constructs except the invisible tasks of type IvT − SS𝑝𝑎𝑟, non-free choice constructs, and IvT − NFC. There is no detection algorithm capable of parallel detection of Short Redo-type invisible tasks and invisible tasks involved in non-free choice constructions.

Figure 4.1. Process discovery algorithms recommendation framework

RECOMMENDATION FRAMEWORK IMPLEMENTATION AND

Implementation in ProM
Evaluation Framework
Evaluation using artificial event logs
Evaluation based on real-life event logs

A screenshot of the implementation of the plugin that records the status of complex builds from the event log. Then, the final recommended algorithm is used to discover a process model from that event log. An illustration of the evaluation method is depicted in Figure 5.3. The comparison of the discovered model with the original model and the event log is performed using conformance checking metrics.

The behavioral precision indicates how much of the behavior of the discovered model is also present in the original model. The behavioral memory reflects how much of the original model's behavior also occurs in the discovered model.

Figure 5.1. A screenshot of the implementation of the plugin detecting the status of complex constructs from event log

HEURISTIC RULE-BASED ALGORITHM

Elaboration of the Ancestor/Descendant Table

For each activity X that appears in an event trace, the following information is extracted from the event log: (i) the name of the activity that immediately precedes activity X (we use 𝑇𝑎(𝑋) to denote the ancestor of X (Predecessor of X)), and (ii) the name of the activity that directly follows activity X (we use 𝑇𝑑(𝑋) to denote the descendant of X (Successor of X)). It consists of 10 activities, and only characteristic event traces representing all possible events of each activity are illustrated. The activity identifier 𝜎𝑖(X,n) indicates the nth occurrence of an activity named X in the event trace 𝜎𝑖.

For example, 𝜎1(B,1) refers to the first occurrence of activity B in 𝜎1, and 𝜎4(B,2) refers to the second occurrence of activity B in 𝜎4.

Table for each activity in each trace. For each activity X that occurs in an event trace, the following information is extracted from the event log: (i) the name of the activity that directly precedes activity X (we use 𝑇 𝑎 (𝑋) to denote

Identification of Standard Constructs

Definition (4) states that if the descendant of task x is equal to y in the track 𝜎𝑖 and the descendant of task y is equal to x in another track and there is no track where the descendants of x and/or y are equivalent to x and y respectively, then the two tasks x and y are parallel. This can be achieved if x is directly followed by y and z and y and z are parallel. In definition (6), the relation (//<) indicates the AND join to synchronize concurrent workflows, and //(x,y)

The relations (

Identification of Complex Constructs

Invisible Task Discovery
Short Loop Discovery
Non-Free Choice Construct (NFC) Discovery
Duplicate Task (DT) Discovery

An invisible task of the long redo type is a task that allows repetition of two or more activities. This invisible task will be placed such that there is an exclusive choice between t and y. An invisible task of type long skip (IvT-LS) is a hidden task that allows two or more tasks to be skipped in execution.

Rule (8) tries to discover an invisible task of type skip that allows the execution of the sequence 𝑥2,. This can happen when there is an invisible task that allows the execution of b and c to be skipped.

Figure 6.1. Sound process models of invisible tasks of type short redo (a) and long redo (b)

Mining Algorithm

Note that duplicate tasks must be detected first so as not to affect the detection of the other constructs. Detected duplicate tasks are logged in tuple θ′ of γ, the renamed task 𝑡′ is added in 𝑇𝑙𝑜𝑔, and the new event log is still reserved in 𝐿−𝐷𝑇. In this chapter, we introduced and explained all the steps of the heuristic rule-based technique for discovering process models from event logs.

For the construction of the heuristic rule-based algorithm, a number of notations have been designed to mine standard constructs (order, AND split/join, and XOR split/join), and a number of heuristic rules have been developed to detect complex constructs. (short loops, invisible tasks, non-free choice constructions and double tasks). One of the most important features of the HR algorithm is that the modularity of the rules usually allows a knowledge base to be easily updated, expanded or modified.

HEURISTIC RULE-BASED ALGORITHM EVALUATION

Evaluation based on small artificial logs
Comparison with other algorithms
Evaluation based on a real-life log
Limitations of the HR algorithm

𝐈𝐯𝐓𝐒𝐒𝐬𝐞𝐪 Yes No Yes Yes Yes Yes No Yes No Yes Yes Yes 𝐈𝐯𝐓𝐒𝐒𝐩𝐫𝐫 Yes No Yes Yes No No No No No No Sb No No No. the HR algorithm . Comparison of mining results by the HR algorithm and current algorithms using a complete real event log.

Using this event log as input, the HR algorithm discovers the N' model shown in Figure 7.4. The correctness of the HR algorithm is based on the assumption that the given event log is complete.

Table 7.1. The performance evaluation results

CONCLUSION

2011)' Process Mining Manifesto' BPM Workshops, Lecture Notes in Business Information Processing, Vol. 2016) 'Study of Detecting and Localizing Draft Drifts from Event Logs in Process Mining' International Journal of Advanced Research in Computing and Communication Engineering, Vol. 2015)' Drift in process mining for the analysis of process changes based on event logs' International Journal of Professional Engineering Studies, Vol.5 No.5, pp.105-110. 2015)’ Concept Drifts Characterization and Detection in Process Mining’ International Journal of Advance Research in Computer Science and Management Studies, Vol.3 No.9, pp.2321-7782.

Lecture Notes in Computer Science, vol Business process mining: An industrial application', Information Systems, Vol. A New Era in Process Mining Tool Support', in Lecture Notes in Computer Science 3536, Ciardo G, Darondeau P (red.), Springer, Berlijn, Heidelberg, pp.