A.J.M.M. Weijters

Business process mining: An industrial application

Contemporary information systems (e.g., WfM, ERP, CRM, SCM, and B2B systems) record business events in so-called event logs. Business process mining takes these logs to discover process, control, data, organizational, and social structures. Although many researchers are developing new and more powerful process mining techniques and software vendors are incorporating these in their software, few of the more advanced process mining techniques have been tested on real-life processes. This paper describes the application of process mining in one of the provincial offices of the Dutch National Public Works Department, responsible for the construction and maintenance of the road and water infrastructure. Using a variety of process mining techniques, we analyzed the processing of invoices sent by the various subcontractors and suppliers from three different perspectives: (1) the process perspective, (2) the organizational perspective, and (3) the case perspective. For this purpose, we used some of the tools developed in the context of the ProM framework. The goal of this paper is to demonstrate the applicability of process mining in general and our algorithms and tools in particular.

Genetic process mining

The topic of process mining has attracted the attention of both researchers and tool vendors in the Business Process Management (BPM) space. The goal of process mining is to discover process models from event logs, i.e., events logged by some information system are used to extract information about activities and their causal relations. Several algorithms have been proposed for process mining. Many of these algorithms cannot deal with concurrency. Other typical problems are the presence of duplicate activities, hidden activities, non-free-choice constructs, etc. In addition, real-life logs contain noise (e.g., exceptions or incorrectly logged events) and are typically incomplete (i.e., the event logs contain only a fragment of all possible behaviors). To tackle these problems we propose a completely new approach based on genetic algorithms. As can be expected, a genetic approach is able to deal with noise and incompleteness. However, it is not easy to represent processes properly in a genetic setting. In this paper, we show a genetic process mining approach using the so-called causal matrix as a representation for individuals. We elaborate on the relation between Petri nets and this representation and show that genetic algorithms can be used to discover Petri net models from event logs.

Process equivalence: comparing two process models based on observed behavior

In various application domains there is a desire to compare process models, e.g., to relate an organization-specific process model to a reference model, to find a web service matching some desired service description, or to compare some normative process model with a process model discovered using process mining techniques. Although many researchers have worked on different notions of equivalence (e.g., trace equivalence, bisimulation, branching bisimulation, etc.), most of the existing notions are not very useful in this context. First of all, most equivalence notions result in a binary answer (i.e., two processes are equivalent or not). This is not very helpful, because, in real-life applications, one needs to differentiate between slightly different models and completely different models. Second, not all parts of a process model are equally important. There may be parts of the process model that are rarely activated while other parts are executed for most process instances. Clearly, these should be considered differently. To address these problems, this paper proposes a completely new way of comparing process models. Rather than directly comparing two models, the process models are compared with respect to some typical behavior. This way we are able to avoid the two problems. Although the results are presented in the context of Petri nets, the approach can be applied to any process modeling language with executable semantics.

A Rule-Based Approach for Process Discovery: Dealing with Noise and Imbalance in Process Logs

Effective information systems require the existence of explicit process models. A completely specified process design needs to be developed in order to enact a given business process. This development is time consuming and often subjective and incomplete. We propose a method that constructs the process model from process log data, by determining the relations between process tasks. To predict these relations, we employ machine learning technique to induce rule sets. These rule sets are induced from simulated process log data generated by varying process characteristics such as noise and log size. Tests reveal that the induced rule sets have a high predictive accuracy on new data. The effects of noise and imbalance of execution priorities during the discovery of the relations between process tasks are also discussed. Knowing the causal, exclusive, and parallel relations, a process model expressed in the Petri net formalism can be built. We illustrate our approach with real world data in a case study.

Process mining based on clustering: a quest for precision

Process mining techniques attempt to extract non-trivial and useful information from event logs recorded by information systems. For example, there are many process mining techniques to automatically discover a process model based on some event log. Most of these algorithms perform well on structured processes with little disturbances. However, in reality it is difficult to determine the scope of a process and typically there are all kinds of disturbances. As a result, process mining techniques produce spaghetti-like models that are difficult to read and that attempt to merge unrelated cases. To address these problems, we use an approach where the event log is clustered iteratively such that each of the resulting clusters corresponds to a coherent set of cases that can be adequately represented by a process model. The approach allows for different clustering and process discovery algorithms. In this paper, we provide a particular clustering algorithm that avoids over-generalization and a process discovery algorithm that is much more robust than the algorithms described in literature [1]. The whole approach has been implemented in ProM.

Genetic process mining: a basic approach and its challenges

One of the aims of process mining is to retrieve a process model from a given event log. However, current techniques have problems when mining processes that contain non-trivial constructs and/or when dealing with the presence of noise in the logs. To overcome these problems, we try to use genetic algorithms to mine process models. The non-trivial constructs are tackled by choosing an internal representation that supports them. The noise problem is naturally tackled by the genetic algorithm because, per definition, these algorithms are robust to noise. The definition of a good fitness measure is the most critical challenge in a genetic approach. This paper presents the current status of our research and the pros and cons of the fitness measure that we used so far. Experiments show that the fitness measure leads to the mining of process models that can reproduce all the behavior in the log, but these mined models may also allow for extra behavior. In short, the current version of the genetic algorithm can already be used to mine process models, but future research is necessary to always ensure that the mined models do not allow for extra behavior. Thus, this paper also discusses some ideas for future research that could ensure that the mined models will always only reflect the behavior in the log.

Behavioural aspects of combining backpropagation learning and self-organizing maps

Backpropagation learning (BP) is known for its serious limitations in generalizing knowledge from certain types of learning material. In this paper, we describe a new learning algorithm, BP-SOM, which overcomes some of these limitations as is shown by its application to four benchmark tasks. BP-SOM is a combination of a multi-layered feedforward network (MFN) trained with BP and Kohonens self-organizing maps (SOMs). During the learning process, hidden-unit activations of the MFN are presented as learning vectors to SOMs trained in parallel. The SOM information is used when updating the connection weights of the MFN in addition to standard error backpropagation. The effect of the augmented error signal is that, during learning, clusters of hiddenunit activation patterns of instances associated with the same class tend to become highly similar. In a number of experiments, BP-SOM is shown (i) to improve generalization performance (i.e. avoid overfitting); (ii) to increase the amount of hidden units that can be pruned without loss of generalization performance and (iii) to provide a means for automatic rule extraction from trained networks. The results are compared with results achieved by two other learning algorithms for MFNs: conventional BP and BP augmented with weight decay. From the experiments and the comparisons, we conclude that the hybrid BP-SOM architecture, in which supervised and unsupervised and learning co-operate in finding adequate hidden-layer representations, successfully combines the advantages of supervised and unsupervised learning.