URL study guide
https://tue.osiris-student.nl/onderwijscatalogus/extern/cursus?cursuscode=2AMI20&collegejaar=2025&taal=enOmschrijving
Understanding and predicting behavior of people and machines in a shared setting (task, project, factory, organization) is central to Data Science and Artificial Intelligence. Actions of people and machines can be recorded as discrete events in event sequences (logs), event databases (tables, graphs), and real-time event streams. Learning behavioral models of discrete event data of human behavior is challenging. Only those events which are causally related may be analyzed together. Further, the analysis results must be fully explainable and interpretable by humans, to evaluate, understand, communicate and improve the model - to let users take correct decisions in concrete situations.This advanced course on process mining teaches students the fundamental concepts and theoretical foundations of process mining along a complete process mining methodology, and exposes students to real-life data sets to understand challenges related to process discovery, conformance checking, and model extension. The course material is based on recent research articles in the field and the course teaches students how to read and understand research literature. The course is organized in two parts:
The common part (weeks 1-3) covers
foundational understanding of event data, constructing new event logs from raw event data, and how to obtain insights into complex event data through pre-processing and data visualization,
foundational event data abstraction and process discovery techniques used in most process mining algorithms, and how different forms of abstraction and algorithmic design decisions impact model quality (accuracy, generalization, and understandability).
This part is concluded by an overview of process prediction techniques where several trace and prefix encoding methods are discussed together with applied machine learning algorithms for prediction in event data.
In the second part (weeks 4-7), students follow one of the following specialization tracks
- Multi-dimensional process mining over event data referring to multiple related objects, resources, actors, and queues. We study how the presence of multiple distinct objects in event data causes classical process mining techniques to generate false and misleading results. We then study how to use -graph-based data models for process mining over multiple related objects. We address data transformation, querying, aggregation, and analysis of complex processes and entire systems using graph databases and study ongoing research and developments in industry such as object-centric process mining..
- (real-time) process mining on event streams where scalable and efficient solutions are leveraged from streaming machine learning to perform online process discovery. Recent online outlier detection and concept drift detection methods for event streams are then presented. The challenges that face online conformance checking are discussed in the context of a state-of-the-art streaming conformance checking method. Then, online prediction models over event streams are presented where we leverage Bayesian models and LSTM-based ones to work for e.g., predicting the next activity under concept drifts and using forgetting mechanisms.
The course is taught as a flipped classroom course with a group-assignment:
- Each topic has a concrete, hands-on reading assignment prior to class; you can use the social reading platform www.perusall.com where you can annotate text passages and discuss with fellow students and lecturers the parts you find difficult to understand.
- We provide for each topic practical exercises: paper-based exercises on the level of the final exam, and tool-based exercises to try out and understand the techniques.
- During class, we discuss applications of the theory through solving assignments and exercises interactively together and thereby specifically address your questions about concepts and ideas you found difficult to understand from reading.
- In the group assignment, you analyze a real-life dataset using a structured process mining and process prediction method using the techniques taught in the course to use process mining to train, evaluate, explain, and improve a prediction model. You specifically learn to critically evaluate behavioral models (descriptive and predictive) regarding their behavioral accuracy and their explainability (generalizability and understandability) and how to use detailed diagnostic information at the level of individual events for model evolution, improvement, and for drawing actionable conclusions towards a stakeholder.
Students must have passed Foundations of Process Mining (2AMI10) to participate as the course 2AMI20 focuses more on advanced topics and ongoing research.
Doelstellingen
After taking this course students should be able to:• have a detailed understanding of the entire process mining spectrum and the methodology for process mining analysis
• can derive and pre-process event logs from raw data and have understand and can work with a specialized form of event data such as event knowledge graphs, or real-time event streams
• have a detailed understanding and be able to explain various concepts for learning descriptive and predictive models from event data and their specific properties and limitations in relation to the event data properties, specifically recognizing temporal patterns and constraints from event sequences, inferring behavioral and causal relations from aggregations of event data, detecting and handling concept drift, outlier detection, conformance checking and stream mining concepts in comparison to static techniques, and the responsible application of these concepts related to fairness, accuracy, confidentiality and transparency
• have a detailed understanding of quality criteria for explainable models and how explainable process models complement general purpose machine learning models, and can evaluate models regarding behavioral accuracy, generalizability, and understandability for event logs as well as multi-dimensional event data or event streams
• independently execute a process mining analysis and critically compare and apply various process mining techniques to discover, evaluate, and improve explainable behavioral models and to draw actionable conclusion