Search

During the last decade several decision mining techniques have been developed to discover the decision perspective of a process from an event log. The increasing number of decision mining techniques raises the importance of evaluating the quality of the discovered decision models and/or decision logic. Currently, the evaluations are limited because of the small amount of available event logs with decision information. To alleviate this limitation, this paper introduces the ‘DataExtend’ technique that allows evaluating and comparing decision-mining techniques with each other, using a sufficient number of event logs and process models to generate evaluation results that are statistically significant. This paper also reports on an initial evaluation using ‘DataExtend’ that involves two techniques to discover decisions, whose results illustrate that the approach can serve the purpose.

Predictive process monitoring is concerned with the analysis of events produced during the execution of a process in order to predict the future state of ongoing cases thereof. Existing techniques in this field are able to predict, at each step of a case, the likelihood that the case will end up in an undesired outcome. These techniques, however, do not take into account what process workers may do with the generated predictions in order to decrease the likelihood of undesired outcomes. This paper proposes a framework for prescriptive process monitoring, which extends predictive process monitoring approaches with the concepts of alarms, interventions, compensations, and mitigation effects. The framework incorporates a parameterized cost model to assess the cost-benefit tradeoffs of applying prescriptive process monitoring in a given setting. The paper also outlines an approach to optimize the generation of alarms given a dataset and a set of cost model parameters. The proposed approach is empirically evaluated using a range of real-life event logs.

The last decade has witnessed an increasing transformation in the design, engineering, and mining of processes, moving from a pure control-flow perspective to more integrated models where also data and decisions are explicitly considered. This calls for methods and techniques able to ascertain the correctness of such integrated models. Differently from previous approaches, which mainly focused on the local interplay between decisions and their corresponding outgoing branches, we introduce a holistic approach to verify the end-to-end soundness of a Petri net-based process model, enriched with case data and decisions. In addition, we present an effective, implemented technique that verifies soundness by translating the input net into a colored Petri net with bounded color sets, on which standard state space analysis techniques are subsequently applied. Experiments on real life illustrate the relevance and applicability in real settings.

Process mining techniques analyze processes based on event data. A crucial assumption for process analysis is that events correspond to occurrences of meaningful activities. Often, low-level events recorded by information systems do not directly correspond to these. Abstraction methods, which provide a mapping from the recorded events to activities recognizable by process workers, are needed. Existing supervised abstraction methods require a full model of the entire process as input and cannot handle noise. This paper proposes a supervised abstraction method based on behavioral activity patterns that capture domain knowledge on the relation between activities and events. Through an alignment between the activity patterns and the low-level event logs an abstracted event log is obtained. Events in the abstracted event log correspond to instantiations of recognizable activities. The method is evaluated with domain experts of a Norwegian hospital using an event log from their digital whiteboard system. The evaluation shows that state-of-the art process mining methods provide valuable insights on the usage of the system when using the abstracted event log, but fail when using the original lower level event log.