Explaining the Impact of Data Characteristics on Process Mining Algorithms
MCML Authors
Andrea Maldonado
Dr.
* Former Member
Abstract
Andrea Maldonado
Dr.
* Former Member
Abstract
This dissertation develops a data-driven framework for robust evaluation in process mining, addressing the lack of standardized benchmarks and reliable assessment methods. It introduces structure-aware data characterization (DROPP, FEEED), bias-mitigating data generation frameworks (GEDI, iGEDI), and explainable experimentation using Shapley-value analysis to quantify how event data characteristics influence algorithm performance. Together, these contributions enable more reproducible, interpretable, and trustworthy evaluation of process mining methods across diverse datasets and scenarios (Shortened.)
BibTeXKey: Mal25