Clustering traces using sequence alignment

Abstract

Process mining discovers process models from event logs. Logs containing heterogeneous sets of traces can lead to complex process models that try to account for very different behaviour in a single model. Trace clustering identifies homogeneous sets of traces within a heterogeneous log and allows for the discovery of multiple, simpler process models. In this paper, we present a trace clustering method based on local alignment of sequences, subsequent multidimensional scaling, and k-means clustering. We describe its implementation and show that its performance compares favourably to state-of-the-art clustering approaches on two evaluation problems.