SECPI: Searching For Explanations For Clustered Process Instances – Review Report

Discussion

Write a Review Report on SECPI: Searching for Explanations for Clustered Process Instances.

Search for Explanations of Clusters of Process Instances or SECPI is a specific technique, which significantly assists the users for understanding a specific trace clustering solution (Van Der Aalst, La Rosa and Santoro 2016). This could be done by simply finding a minimum set of control flow characteristics. The absence of these control flow characteristics could eventually prevent the process instance from remaining within the current or recent cluster. The shortcoming of the current trace clustering technique about the insight provisioning into the proper computation of the partitioning is subsequently addressed by knowing concise individual regulations, which could easily explain the reason of the certain instance as the part of the cluster (Jeston 2014). The partitioning event logs within the several groups of process instances are the most convenient and basic recipe to address the various challenges for dealing with the complicated event logs. These event logs present the larger amount of the process behaviour.

The following report reviews on the article of SECPI: Searching for Explanations for Clustered Process Instances by authors Jochen De Weerdt and Seppe vanden Broucke. The report will be properly reviewing this paper properly. This review report will be explaining the entire concept of SECPI with respect to process discovery, trace clustering, user comprehension, support vector machines and instance level explanations. Relevant justifications are provided within the review report for understanding the entire concept properly and perfectly.

Trace Clustering

Trace clustering can be defined as the significant approach for dealing with the various problems that most of the event logs comprise of the extensive amount of the distinct behaviours (Di Francescomarino et al. 2016). This distinct behaviour is the process variant, since it enables the specific user in splitting up of log so that the various distinct models could be eventually learnt for the purpose of describing the various underlying business processes.

Two specific groups of collections of the approaches for trace clustering could be easily discerned with over the one hand technique, relying on the principle of the distance based clustering. The other hand technique helps to incorporate the particular model driven approach (Pan et al. 2013). The very first group comprises of the various tools and techniques that transform the input event log to any propositional format. This is done for applying all the well known clustering techniques from the domain of data mining.

Trace Clustering

Business process management or BPM is the typical discipline within the operations management, which utilizes the several methods for discovering, modelling, analyzing, and measurement, improving, optimizing and finally automating the several business processes (Dumas et al. 2013). This BPM eventually focuses on the improvement of the corporate performance by simple management of the business processes. The proper combination of the methods s utilized for the management of the business process of the company. All of these business processes could be structured as well as repeatable or could be unstructured as well as variable. There are various technologies that are enabled with the utilization of this business process management or BPM. This type of process management is different from the program management; since program management is concerned with the management of the group of various inter dependent projects. On the other hand, the process management involves program management within it (Rosemann and vom Brocke 2015). The process management is the perfect utilization of the repeatable process that helps in the improvement of the result of the project.

This BPM or business process management helps to improve the business processes from one end to the other by simply analyzing it properly and thus modelling the procedure of the scenario of different scenarios for the execution of the improvements and monitoring all the improved process for continuous optimization (Hammer 2015). Trace clustering technique plays the most significant role in the business process management since relevant information could be easily extracted for the purpose of perfect information utilization. For the well structured processes such as work flows, discovery feature of the process mining has restricted the appeal, as it is bound to confirm the fact that the prescribed features of this process are perfect (Van der Aalst 2013). Although, most of the real life business procedures are not at all strictly enforced by the supporting and existing information system, there are some of the major notions of these processes.

The most significant problem or issue with the existing techniques for trace clustering is that they are responsible for providing extremely less or almost zero insight within the actual reasoning of the partitioning any event log in the proper method (Chang 2016). From the model learning perspectives, this clustering bias of the technique of trace clustering helps to determine the procedure of any solution being constructed. All of these clustering techniques that are described within the literature of process mining have the ability in employing the broad variety of various clustering biases. On one hand, the subset of the techniques subsequently relies on the entire concept of distance as the proper measure of the instance similarity (Trkman 2013). Each and every model driven technique relies on the maximum likelihood as well as fitness optimization.

Business Process Management

For the distance based clustering, the various techniques for data mining like hierarchical clustering or k means are properly applied. The distance is the potential candidate for the proper explanation of the clustering result (Becker, Kugeler and Rosemann 2013). One can easily visualize the several instances within the networked graph and thus making use of the comparative statistical analysis of every underlying variable, which could determine the intra cluster distance and the inter cluster distance. There is a projection of the process instances onto the process features that would typically generate the large amount of variables. This would eventually make the situation extremely complicated approach. To this, it should be added that due to the large amount of variables, distance-based techniques suffer from the curse of dimensionality problem. The conventional proximity metrics in the higher dimensional space might not be qualitatively meaningful (De Weerdt et al. 2013). Hence, it is argued that the entire value of the distance concept to assist the users with proper understanding of the lower trace clustering solution. A lot of expertise is required in this process and thus should be checked subsequently. Trace clustering for business process management or BPM is the most effective of all and hence is termed as the best of all.  Moreover, the utilization of this particular technique is much higher than any other techniques (Song et al. 2013). SECPI or the Searching for Explanations for Clustered Process Instances is the most significant technique of the trace clustering technique and thus should be utilized properly. Most of the users opt for the trace clustering as the generalization and precision is checked in the process.

Instance Level Explanations with SECPI

This paper has described the complete new analysis approach to explain all the differences within the clusters of the process instances (Sadiq, Soffer and Völzer 2014). The major idea of this particular approach is that instead of providing the global explanation, a short and precise rule are considered for every instance of individual. This is done by joining the control flow characteristics, for the purpose to form the antecedent and cluster switch as the consequences. The most significant objective of this technique is that accurate as well as concise explanations are learnt with the help of this particular technique. Moreover, various factors are determined in this process.

There are some of the major processes or steps in this specific technique of SECPI or Searching for Explanations for Clustered Process Instances. The first and the foremost step in this technique is the construction of the data set. At first, all the process instances are being converted to various feature vectors. This implementation thus supports the various attribute templates (van Oirschot et al. 2014). However, the initial experiments demonstrate that direct follow attribute template has the ability in providing perfect explanatory power from the perspective of control flow. The optimal configuration of this featurisation step is investigated in this manner. Hence, the labelled set of data is properly obtained in which the supervised techniques for data mining could be easily applied.

Instance Level Explanations with SECPI

The second step in this technique of SECPI or Searching for Explanations for Clustered Process Instances is properly deriving explanations from the support vector machine or SVM classifier (Sadiq, Soffer and Völzer 2014). The approach of SECPI or Searching for Explanations for Clustered Process Instances is eventually inspired a specific algorithm that help to document the classifications. The most significant similarity is the utilization of the SVM based classifier since the base model from where all the explanations are eventually derived. As for document classification, SVMs are ideally suited in our context because the use of multiple or complex attribute templates will quickly lead to massive dimensionality. By proper employment of the well known large-scale linear classification based on linear kernel SVMs, the approach could be supporting the data with several features as well as instances (Sadiq, Soffer and Völzer 2014). The most significant contribution of this particular paper comprises of the adaptation of the approach for the trace clustering with various modifications. In the beginning, supporting the middle class prediction is being developed since this context is more plausible for over two clusters. The next step is the configuration of the algorithm where all the explanations could be limited to the behaviour that is present within the process instance (Di Francescomarino et al. 2016). The third significant step is that various performance optimizations had been introduced. The attributes are avoided with zero variability and thus the prevention of the repetitive checking of the similar attribute combinations and the proper expansion of attribute combinations is checked and considered.

A formalized preview of the algorithm of SECPI or Searching for Explanations for Clustered Process Instances is provided in the paper. Proper inputs are explained in this algorithm. Various binary attributes are generated with the help of attribute templates. In the second step of the algorithm, a classifier is trained on the set of data for any provided featured vector and a predicted class label. After the perfect explanation of the algorithm, a proper set of attributes are obtained (Sadiq, Soffer and Völzer 2014). All the indices set represent the candidate explanation and thus proper interpretation is required. The SVM model with the highest probability determines the label. SVMs remain more suitable classifier for using within the proposed technique of SECPI in business process management.

Conclusion

Therefore, from the above review report, conclusion can be drawn that SECPI or Search for Explanations of Clusters of Process Instances is the brand new technique that is properly assisting the users with the proper understanding of the results of trace clustering. This paper has perfectly explained the concept of Search for Explanations of Clusters of Process Instances. The requirement for such a technique eventually stems from the core observation that the specific technique for trace clustering does not give adequate insight into the procedure that how this clustering solution is being composed. In the future work, it is seen that there is expansion on several number of intensely related topics. In the beginning, the impact of the attribute templates that are utilized is planned to be inspected, since they play the most significant role in representation of the control flow domain. Moreover, the proper investigation of the incorporation of the non control flow based attributes. The second point is that the aggregation of the instance level explanations is the most significant research track in this review report. The recent implementation subsequently supports the basic investigation of various shared explanations within the instance groups. The more intelligent visualization techniques and the rule clustering techniques are investigated in this purpose. The various practical use cases are focused where the SECPI is proving extremely beneficial. The significant discovery of the process model collection that is user driver is the significant area where the feedback mechanism is supported. Moreover, the technique of SECPI has the capability to relate to the exogenously defined clusters like the higher versus lower cost instances and process specific control flow characteristic. The above review report has properly provided the review analysis of the paper by the authors of Jochen De Weerdt and Seppe vanden Broucke.

References

Becker, J., Kugeler, M. and Rosemann, M. eds., 2013. Process management: a guide for the design of business processes. Springer Science & Business Media.

Chang, J.F., 2016. Business process management systems: strategy and implementation. CRC Press.

De Weerdt, J., vanden Broucke, S., Vanthienen, J. and Baesens, B., 2013. Active trace clustering for improved process discovery. IEEE Transactions on Knowledge and Data Engineering, 25(12), pp.2708-2720.

Di Francescomarino, C., Dumas, M., Maggi, F.M. and Teinemaa, I., 2016. Clustering-based predictive process monitoring. IEEE Transactions on Services Computing.

Dumas, M., La Rosa, M., Mendling, J. and Reijers, H.A., 2013. Fundamentals of business process management (Vol. 1, p. 2). Heidelberg: Springer.

Ha, Q.T., Bui, H.N. and Nguyen, T.T., 2016, September. A trace clustering solution based on using the distance graph model. In International Conference on Computational Collective Intelligence (pp. 313-322). Springer, Cham.

Hammer, M., 2015. What is business process management?. In Handbook on Business Process Management 1 (pp. 3-16). Springer, Berlin, Heidelberg.

Jeston, J., 2014. Business process management. Routledge.

Pan, G., Qi, G., Zhang, W., Li, S., Wu, Z. and Yang, L.T., 2013. Trace analysis and mining for smart cities: issues, methods, and applications. IEEE Communications Magazine, 51(6), pp.120-126.

Rosemann, M. and vom Brocke, J., 2015. The six core elements of business process management. In Handbook on business process management 1 (pp. 105-122). Springer Berlin Heidelberg.

Sadiq, S., Soffer, P. and Völzer, H. eds., 2014. Business Process Management: 12th International Conference, BPM 2014, Haifa, Israel, September 7-11, 2014, Proceedings (Vol. 8659). Springer.

Song, M., Yang, H., Siadat, S.H. and Pechenizkiy, M., 2013. A comparative study of dimensionality reduction techniques to enhance trace clustering performances. Expert Systems with Applications, 40(9), pp.3722-3737.

Trkman, P., 2013. Increasing process orientation with business process management: Critical practices’. International journal of information management, 33(1), pp.48-60.

Van der Aalst, W.M., 2013. Business process management: a comprehensive survey. ISRN Software Engineering, 2013.

Van Der Aalst, W.M., La Rosa, M. and Santoro, F.M., 2016. Business process management.

van Oirschot, Y.P.J.M., van Dongen, B.F., Buijs, J.C.A.M. and Dijkman, R.M., 2014. Using Trace Clustering for Configurable Process Discovery Explained by Event Log Data (Doctoral dissertation, Master’s thesis).