Mining complex business processes for compliance checking and end-state prediction

Abstract

Real-world business processes are dynamic, with unstructured event logs including various business classes. These processes are critical to the quality of services provided to customers; thus, businesses need advanced warning if an executing process is not progressing as expected. However, business processes in the complex real-world context are varied and challenging to monitor for any disparities. Exploring the sources of errors in business process execution is one of the most common and complex challenges for today’s enterprises. Abnormalities are typically characterised by the lack of essential aspects of a process or the presence of undesirable behaviour during process execution.

Process mining is a fusion of data mining and business process modelling, which provides a scientific foundation for analysing process data execution of business processes. Process mining utilises data from event logs to discover, monitor, and enhance the quality of processes. A process model could be used to analyse and anticipate future process paths and identify anomalous execution. Process mining is inevitable in today’s corporate environment since business process data is complicated and contains various information about the behaviour of instances. Conformance techniques in process mining provide real-time information on how a process instance is executing.

Though process mining examines event logs to improve process execution, large and complex event logs make traditional process mining methodologies challenging to apply. Techniques like process discovery lead to Spaghetti models, which are difficult to mine. To maintain the quality of business processes, a system that analyses and gives constructive feedback for process improvement is necessary. Process mining techniques extract useful information from large and complex event logs, but it’s crucial to translate them into simple and logical segments. Process mining approaches tend to improve processes but utilising event logs for anomalies detection is still a new field of research.

There are techniques in the literature that use association rules and process model-based features to anticipate a process instance’s future outcome. However, they lack empirical knowledge of an instance’s behaviour relative to ideal process conduct. In real-world business contexts, trace clustering techniques are widely used in literature to analyse heterogeneous process executions, yet event-level clustering techniques are not intensively investigated. A combination of trace clustering and event clustering would simplify complex business logs, and clustering outcomes should be easily understandable to business users.

This thesis attempts to determine strategies for analysing real-world business processes for monitoring, abnormality detection, and root cause analysis. For this, several techniques are proposed in this work, including a novel type of process termed ‘Contrail’, which genuinely represents the behaviour of execution of cases in the real-world business environment. Two new techniques for clustering of event log are proposed. A large complex business process log is decomposed into sub-logs without using event abstraction through the Common Events Identifier (CEI) algorithm. Another technique is Novel Hierarchical Clustering (NoHiC), a multi-stage framework for business-logic driven clustering of highly variable process logs. The discovered clusters provide critical business context information that assists business users in comprehending how their organisational operations are carried out.

Furthermore, a novel technique for identifying abnormalities in business process execution is presented in this thesis through the extension of available conformance analysis techniques. The work is concluded with root cause analysis to pinpoint the abnormalities in the process execution. All the techniques are applied to a real-world Customer Relationship Management (CRM) process of UK’s renowned telecommunication firm’s logs.

Finally, the thesis identifies prospective study areas of interest. Future studies could compare how the same processes are executed in different geographical areas and related organisations with varying levels of competence and resources. This research can also be applied to other business areas such as healthcare, production and manufacturing, project management, higher education, and agriculture.

Date of Award	Nov 2022
Original language	English
Sponsors	BT Ireland Innovation Centre (BTIIC) & Invest Northern Ireland
Supervisor	Ian Mc Chesney (Supervisor), Sally McClean (Supervisor), Darryl Charles (Supervisor) & Paul Taylor (Supervisor)