부산대학교 도서관

Process mining uses historical executions of business processes (as recorded in an event log) to uncover and describe the process’ behaviour as a process model. The goal of conformance checking is to ensure that the model is of high quality and is a good representation of the event log, thus assuring the model is a solid foundation for subsequent analysis of the process. To date, few conformance checking approaches have considered model quality beyond what is determined by the execution order of process activities. In data-aware process models, which are generated by decision mining techniques, process activities are annotated with conditions to represent the decision-making of a process. Such models are more expressive than those that represent only process activities. With the current notions of conformance checking, it is unclear what properties determine the quality of these data-aware process models. To address this gap, we introduce desirable properties, as axioms, for conformance checking of data-aware models. Our contribution is threefold: i) we present a generalisation that abstracts from the representation of data-aware models, ii) we present nine axioms of desirable properties for data-aware conformance checking, and iii) we define two measures for model recall and precision. Using our axioms as a yardstick, we compare our proposed recall and precision measures with existing measures. Our experimental results show that existing measures exhibit limited adherence to our axioms; while, our two proposed measures exhibit high adherence to our axioms.