Multi-Perspective Clustering of Process Execution Traces

Abstract

Process mining techniques enable extracting process models from process event logs. Problems can arise if process mining is applied to event logs of flexible processes that are extremely heterogeneous. Here, trace clustering can be used to reduce the complexity of logs. Common techniques use isolated criteria such as activity profiles for clustering. Especially in flexible environments, however, additional data attributes stored in event logs are a source of unused knowledge for trace clustering. In this paper, we present a multi-perspective trace clustering approach that improves the homogeneity of trace subsets. Our approach provides an integrated definition of similarity between traces by defining a distance measure that combines information about executed activities, performing resources, and data values. The evaluation with real-life event logs, one from a hospital and one with traffic fine data, shows that the homogeneity of the resulting clusters can be significantly improved compared to existing techniques.