Master Thesis - Context-aware detection of deviations in process executions
Process mining techniques enable practitioners to have better understandings of their business processes. Major categories of those techniques include 1) process discovery that derives process models from historical records of the process execution, 2) conformance checking that finds deviations between the reference process model and the real execution of the process, and 3) enhancement that identifies bottlenecks in the process with the process model enriched with performance measures. These are considered as backward-looking techniques aiming at finding insights regarding the past execution of the process.
Recently, the focus of process mining field has been shifted into forward-looking techniques from which process managers can identify insights that can be used to influence the future of running process instances. Deviation detection is one of those techniques. It aims to identify the deviations in business processes that negatively affect the performance of them. It has many application domains, such as detecting fraudulent claims in insurance companies.
Deviation detection has been actively studied in the field of process mining. It can be categorized into three major categories, i.e., 1) model-based approach, 2) clustering-based approach, 3) profile-based approach. The model-based approach employs conformance checking techniques on a discovered process model to detect deviations. The clustering-based approach basically deploys clustering methods and consider the cases in small clusters as deviations. The profile-based approach identifies mainstream cases and classifies other cases that are not similar to them as deviations.
There are two missing links which are essential, but not efficiently managed by existing techniques. First, contexts play a key role in deviation detection since the same behaviors can be seen differently according to their contexts. Let’s imagine an e-commerce business like Amazon. The same delivery times (e.g., three days taken for two different orders) can have different meanings depending on the situations (e.g., delivery in Black Friday with excessive demands vs. regular weekdays). Second, multi perspectives should be considered when defining deviations. For instance, the control-flow perspective (e.g., the existence of a task) must be considered along with the organizational perspective (e.g., the employee). Suppose the task, “Acquire approval”, must be preceded for “changing the priority of a customer”. Any violation of this control-flow is a deviation. However, if an employee at the manager level is in charge of changing priorities, then it is considered okay to skip “Acquire approval”.
In this Master thesis, the student is required to develop novel multi-perspective and context-aware deviation detection techniques that take object-centric event log and contextual information such as organizational structures as inputs, and deviated process instances considering different contexts and multi-perspectives as outputs. This includes the compact and concise formalization of the techniques, the implementation of the techniques in Python packages, and the comprehensive evaluation of the techniques using extensive real-life data sets.
- Data Science in Action
- A framework for detecting deviations in complex event logs
- Balanced multi-perspective checking of process conformance
- Object-Centric Process Mining: Dealing With Divergence and Convergence in Event Data
- Basic computer science concepts
- An interest in process mining
- Proficient in python with experience in object-oriented programming.
For more Information
Send an e-mail to Gyunam Park. Make sure to include a C.V., detailed information about your background, and scores for completed courses.