Improving Precision in Process Trees Using Subprocess Tree Logs

Process mining is a family of techniques that provide tools for gaining insights from processes in, for example, business, industrial, healthcare and administrative settings. Process discovery, as a field of process mining, aims to give a process model that describes a process given by an event log. A process model describes an underlying process well if it contains all behavior relevant (fitness) and if it does not model behavior that is not contained in the event log (precision). The Inductive Miner (IM) family provides algorithms to find process models on complex event logs efficiently and in an easy-to-understand process model representation using process trees. Due to its characteristics, the IM family is one of the state-of-the-art discovery algorithms and is implemented in software of market-leading process mining vendors. Nevertheless, process trees and in particular those discovered by the IM can have imprecise parts. In this work, we combine existing work and present an approach that replaces such parts with more precise parts while preserving fitness. In addition, we demonstrate the frameworks applicability and utilization by improving process trees discovered by the IM, using the IM itself. Further, guarantees on the preservation of fitness and precision are given. Our experiments clearly show that our techniques can be applied to real-life event logs and that they lead to an improvement in precision.