A ground truth approach for assessing process mining techniques.

Area of Science:

• Computer Science
• Information Systems

Background:

• Assessing process mining techniques with real-world data is challenging due to missing ground truth and data imperfections.
• Existing synthetic data generation methods often fail to capture realistic behavioral deviations and lose the model-log link.

Purpose of the Study:

• To propose a ground-truth approach for generating realistic synthetic process data, including behavioral deviations and recording errors.
• To enable robust evaluation of process mining techniques by providing ground truth knowledge.

Main Methods:

• Generating synthetic process data from initial process models (automatic or hand-made).
• Incorporating patterns of behavioral deviations and recording errors to create deviating models and imperfect event logs.
• Utilizing the initial model, deviating model, and imperfect log for assessment.

Main Results:

• Demonstrated the approach on three synthetic process datasets.
• Applied the generated data in a conformance checking use case to assess systemic alignments.
• Showcased the ability to expose and explain deviations between modeled and recorded behavior.

Conclusions:

• The proposed ground-truth approach provides superior quantitative and qualitative insights into process mining technique performance compared to traditional methods.
• This method enhances the understanding of strengths and weaknesses of process mining techniques by simulating realistic imperfections.