Access Type

Open Access Dissertation

Date of Award

January 2021

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Industrial and Manufacturing Engineering

First Advisor

Ekrem Alper Murat

Abstract

Process Mining (PM) is a new era in business development management that reinforces business process sustainability. Process Mining uses various techniques to discover the process model and identify the root cause analysis of process delays based on the event log. There are three main types of process models: procedural models, declarative models, and hybrid models. Procedural models tend to discover the main pattern of the activity flows in the process. In contrast, declarative models analyze the process behavior and express this behavior as a compact set of rules between pair of two activities. Hybrid models are a combination of procedural and declarative models. However, these models suffer from inconsistency, in that they do not discover the full picture of the process since they rely on primarily event log input. While extensive research has been conducted to discover process models from event logs, methods to integrate the domain expert user’s knowledge into the discovery has been lacking. In many cases, the discovered process model from the event log contrasts with the process model discovered by user knowledge. This dissertation aims at filling this gap by developing methods to integrate user’s process knowledge within the event log driven process discovery for both declarative and procedural models. This dissertation has a two part contribution: (1) An integrated declarative process discovery approach which uses predictive modeling and simulated user’s knowledge representation. In this approach, we first construct an ensemble predictive model to identify the type of declarative constraints between pairs of activities based on two independent determinations (i.e., user and event-log generated determinations) of the constraint and their associated features. To capture user’s process knowledge, we developed a simulation based optimization approach which first generates a large set of simulated traces per the user input and then optimally selects a subset of traces. Results using an experimental dataset indicates that this integrated approach is able to achieve 82 \% accuracy in determining constraint types between pairs of activities. (2) A sampling based optimization approach to discover procedural models with better conformance using user input. This approach first discovers a process model using existing event-log based approaches and then improves its conformance by repetitively sampling user informed trace sets and refining the model. Results show that this approach can effectively improve the conformance of the discovered process for the collective process information available from the event logs and user knowledge. This work contributes to the domain of process mining by developing a method to capture the process knowledge of a domain expert user, a predictive method for declarative constraint identification with domain expert user input, and an optimization approach to enhance the conformance of discovered process model using domain expert user’s process knowledge.

Download

Share

COinS