Enabling Variability Mining in Large Organizations through Event Log Collections Simplification

This thesis addresses the problem of managing Variability in large collections of event logs underlying families of processes. Variability occurs when processes, sharing the same purpose, are executed in slightly different ways. A simplified representation of the family offers significant advantages, such as the ability to make changes effectively, create opportunities for reuse, and analyze the collection more simply. This thesis proposes methodologies that are entirely data-driven, aligned with the sig- nificant relevance that process mining and data usage have in the Process Science community. One of the challenges in managing process families is their complexity, which becomes even more relevant in the case of large event log collections. This thesis aims to create methodologies that simplify the collection by recognizing variants through clustering methodologies and allowing the identification of outdated behavior within them. The methodologies are enriched with visualizations that align with the growing sensitivity of the community towards visualization, respecting one of the foundational values of the Process Science community, which is to facilitate communication among stakeholders. This thesis follows the principles of Design Science Research Methodology and, in line with them, proposes various methodologies tested in a real scenario, the Public Admin- istration. This is a natural laboratory for observing the phenomenon of Variability, by virtue of the need that processes in this domain must satisfy, i.e. simultaneously obey national and general laws and specific requisite required by local laws. Five methodologies are proposed that address specific problems, such as the clustering of large collections and the identification of concept drift phenomena generated by the evolution of collections over time. A sixth methodology is also presented that unifies the previous ones, showing how the combined application of the proposed approaches results in an effective simplification of the Event Log collections.