Work

A Deep Learning Approach to Business Process Mining

Year: 2023
Type: dissertation
Abstract: Competing and evolving markets force organisations to continuously monitor, evaluate, and optimise their business processes. To do the task at scale, organisations often turn to automatic mining of process execution logs constantly generated by various information systems. Many open-source and commercial tools have been developed in recent years to help organisations perform various process mining tasks using process execution logs (often called event logs), such as process discovery, conformance checking, and detecting drifts in processes. Compared to traditional process mining techniques such as Petri nets and Business Process Model and Notation (BPMN), deep learning methods such as Recurrent Neural Networks and Long Short-Term Memory (LSTM) in particular have proven to achieve better performance in terms of accuracy and generalising ability when predicting sequences of activities performed as part of business processes based on event logs. However, unlike traditional network-based process mining techniques that can be used to visually present all activity sequences of the discovered business process, existing deep learning-based methods for process mining lack a mechanism explaining how the activity sequence predictions are made. To address this limitation, this thesis proposes an extensible process mining solution that combines the benefits of interpretable graph-based methods and more accurate but implicit deep learning methods. The main contributions of this research are: (i) building an LSTM model for predicting business process activity sequences from event logs that outperforms existing state-of-the-art deep learning solutions; (ii) proposing a graph-based approach to explaining the decision-making process of the LSTM model when predicting business process activity sequences; and (iii) developing methods for detecting and localising sudden concept drift in event logs (i.e., offline) and event streams (i.e., online) using deep learning and graph-based approaches. The proposed methods have been extensively evaluated by conducting experiments using real-life and artificial event logs and have been demonstrated to outperform existing state-of-the-art solutions in many cases.
Language: English
Open Access status: green
OpenAlex ID: W7027786916
Cites: 110
Cited by: 3
Related to: 10