On decomposed process discovery

: how to solve a jigsaw puzzle with friends

Abstract

Process mining aims to discover, monitor and improve real processes by extracting knowledge from event logs readily available in today's information systems. Event data provide the starting point for process mining. Such data can be found in any information system. Because of the ever increasing number of information systems which produce event data and more and more event (meta-)data being recorded, event logs grow bigger and bigger. Big Data is a recently coined term used to describe the enormity of data available in today's information systems. As a result of the exponential increase in event data, existing process mining techniques have trouble dealing with today's needs. Decomposed process mining aims to solve this problem by decomposing the process mining problem into many smaller problems which can be solved in less time, using less resources, or even in parallel. Many decomposed process mining techniques have been proposed in literature, the latest of which is a generic approach. This thesis discusses this generic decomposed process mining approach in detail, focusing on decomposed process discovery. The approach is analyzed using a series of tests. Results show that even though the decomposition step takes a relatively small amount of time, it is of key importance in finding a high-quality model and for the computation time required to discover it. Currently there is only one way to decompose an event log, which is shown to be too finegrained. We define three quality notions that can be used to assess a decomposition upon, before using it to discover a model or check conformance with. We then propose an agglomerative hierarchical approach which is able to find a high-quality decomposition in little time. This approach is compared to the existing approach through a series of tests. All work presented in this thesis has been implemented in the process mining tool ProM, and is available to the public.

Date of Award	30 Jun 2014
Original language	English
Supervisor	W.M.P. van der Aalst (Supervisor 1), H.M.W. (Eric) Verbeek (Supervisor 2) & I.T.P. Vanderfeesten (Supervisor 2)