Alignment Approximation using Subsets of Modeled Behavior
BSc Thesis Project
Title: Alignment Approximation using Subsets of Modeled Behavior
Author: Julian P. Schnitzler
Supervisor: Sebastiaan J. van Zelst
1st Examiner: Prof. Dr. ir. Wil M.P. van der Aalst
2nd Examiner: Prof.Dr.rer.nat. Peter Rossmanith
Summary
Process Mining builds a bridge between the fields of Process Science and Data Science.Data generated by a process is often covered in an Event log. One of the main sub-fields of Process Mining is Conformance Checking, which aims to evaluate the quality of a process model compared to an Event log. In Conformance Checking, computing alignments is a common approach to compare a log trace and a model. As the calculation of alignments can be time consuming, various approximation techniques exist. Recently, edit-distance based techniques were introduced. While these techniques approximate the alignment cost, they do not provide actual alignments. Multiple approaches exist that take traces generated from a model, and compare one or multiple model traces to a log trace. In this thesis, we propose a method to compare multiple model traces at the same time to a log trace to receive an alignment approximation. Also, we show compression techniques to improve our approach in terms of runtime. We evaluate our method on two real-life event logs. The results indicate that improvements are possible, and depending on the shortest path algorithm either comparing all traces at once to the log trace or comparing only one trace at a time and taking the best result can be beneficial.