• No results found

Evaluation of Use Case 1 (Structure of the process)

4.4 Conclusion

5.1.2 Evaluation of Use Case 1 (Structure of the process)

Description: Determine the structure of an unknown process or discover how a process looks like in practice.

Sub use case 2: XOR splits and joins

Acceptance criteria: The tools supporting this sub use case provide a control flow representation that suggests that after the completion of an activity that is considered to be a XOR split, the thread of control is passed to exactly one of the outgoing branches. The convergence of the multiple branches into one subsequent branch is, then, done by executing an activity, considered to be a XOR join.

Event log used: Event log E2 (Table E.2). The event log consists of activity A, followed by the possi- bility of executing one of the activities B or C, and activity D at the end.

ARIS PPM The XOR split pattern can be discovered by both mining approaches implemented in ARIS PPM (EPC and activity sequence). Figure 5.1 shows the EPC process model. Therefore, the system supports the XOR splits and joins sub use case.

Futura Reflect The XOR splits and joins sub use case is supported by both Mine and Explore process discovery functionalities. Both discovered models (Figures 5.2(a) and 5.2(b)) depict an exclusive choice construct after activity A.

ProcessAnalyzer The process model represented in Figure 5.3 contains an exclusive choice pattern after activity A. The XOR splits and joins sub use case is thus supported by ProcessAnalyzer.

5.1. USE CASES-BASED EVALUATIONS 35

Figure 5.1: ARIS PPM supports the discovery of XOR splits and joins

(a) XOR split - Futura Reflect - Mine (b) XOR split - Futura Reflect - Ex- plore

Figure 5.2: Futura Reflect supports the discovery of XOR splits and joins in both miners

36 CHAPTER 5. EVALUATION RESULTS

Sub use case 4: Parallel activities

Acceptance criteria: The tools supporting this sub use case provide a control flow representation sug- gesting the divergence of a branch into multiple branches that execute concurrently.

Event log used: Event log E4 (Table E.4). The event log consists of traces starting with activity A, followed by the sequences B-C or C-B (activities B and C are executed in parallel, therefore), ending with activity D.

ARIS PPM The discovery of Parallel activities has a limited support in ARIS PPM in the EPC repre- sentation (Figure 5.4). The limited support is due to the condition that the activities involved in the concurrent construct need to have the same start and completion timestamps. The representation using activity sequences does not support this sub use case.

Figure 5.4: ARIS PPM provides limited support for the discovery of parallel activities in the EPC repre- sentation due to the fact that activities B and C need to have identical start and completion timestamps Futura Reflect The Parallel activities sub use case is supported by Mine, but not by Explore. The

model built by the Explore miner (Figure 5.5(b)) shows no parallelism, but a combination of the two traces in the input event log. The algorithm used by Mine is, however, able to detect that activities B and C can be performed in parallel (Figure 5.5(a)).The use of the special “+” symbol in the boxes representing activities A and D denote AND split, respectively AND join, semantics. ProcessAnalyzer The model discovered (Figure 5.6) contains no representation that might suggest that

the execution of activities B and C is concurrent in practice. Instead, the model covers the two possible paths from the event log (A-B-C-D and A-C-B-D) by displaying the exact two sequences. Moreover, the model allows for additional behavior (e.g. A-C-D, A-B-D, etc), not observed in the log. Therefore, the Parallel activities sub use case is not supported.

Discovering parallel construct is not an easy task. One of the challenges associated to it is the ex- plosion of possible tasks combination, mentioned in [8, 4, 36]. For the use-case based evaluation, we

5.1. USE CASES-BASED EVALUATIONS 37

(a) Parallelism - Futura Reflect - Mine

(b) Parallelism - Futura Reflect - Ex- plore

Figure 5.5: Futura Reflect supports the discovery of parallel activities in the Mine miner, but not in the Exploreminer

Figure 5.6: ProcessAnalyzer does not support the discovery of parallel activities

tested the ability of the tools of discovering parallelism using a quite simple event log with only one level of concurrency involving only two activities. In this situation, even the models that do not contain the parallel construct, are readable. However, in most of the real cases, processes are far from being that simple and the inability of representing concurrency can become an issue with respect to the readability and clarity of a process model.

In the following, we illustrate this problem by showing the models discovered based on the highly parallel process, described in Section 4.1.3. The CPN tool representation of the initial process is depicted in Figure 5.7. The activities in the process have random durations and the executions of the activities involved in a parallel constructs overlap in time.

The models depicted in Figures 5.8 and 5.9 are the results of algorithms unable to detect parallelism. The first model was obtained by using ProcessAnalyzer and the second one is the result of the Explore miner from Futura Reflect. The two models are identical. Opposite to this, we present the model dis- covered using the Mine functionality in Futura Reflect in Figure 5.10. The model correctly captures the situations in which activities are executed concurrently. The difference between the structure, the com- plexity, and the understandability of the the two sequential models and the parallel one proves the fact that the inability of discovering and representing parallelism has a negative impact on the readability of a process model.

Due to the random activities durations, the condition required by ARIS PPM for activities to have the same start and end timestamps does not hold. As a result, the system could not reproduce the parallel

38 CHAPTER 5. EVALUATION RESULTS

Figure 5.7: Representation in CPN tools of a highly parallel process model with random activity dura- tions

constructs present in the initial process. However, we modified the process by introducing constant ac- tivities durations values to enforce identical start and completion timestamps for activities. The resulting groups of activities that meet this condition are B-C-D and E-F-I-H. The corresponding event log was mined using ARIS PPM and the result is depicted in Figure 5.11. The discovered model is correct given the groups of activities for which executions completely overlap in time.

Figure 5.8: ProcessAnalyzer is unable to detect the parallel constructs present in the highly parallel process

5.1. USE CASES-BASED EVALUATIONS 39

Figure 5.9: The Explore miner in Futura Reflect is unable to detect the parallel constructs present in the highly parallel process

Figure 5.10: The Mine miner in Futura Reflect is able to correctly capture the parallel constructs present in the highly parallel process

40 CHAPTER 5. EVALUATION RESULTS

Figure 5.11: ARIS PPM is able to discover parallelism if the concurrent activities have identical start and end timestamps