Variants of Variants: Context-Based Variant Analysis for Process Mining

An essential aspect of analyzing processes with process mining is the notion of variants. Analysts can make better decisions and improve processes by understanding variants based on event data. Conventional variant analysis methods, however, focus primarily on variation based on event sequences only, which means they cannot effectively capture the complexity and contextual specificity of real-world process variation in other dimensions. In addition, the few methods that allow for a multifaceted analysis presume process variants as already predefined entities rather than customizable to specific target use cases. To overcome this limitation, we reconceptualize variant analysis at the level of variant extraction and propose an approach that maps process instances to equivalence classes over a set of properties utilizing an unsupervised binary mapping technique. The approach was implemented as a Python application, allowing analysts to define various variant types flexibly. We validate the approach by creating domain-specific variants with real-world event data representing pathways of sepsis patients from a hospital. The evaluation demonstrates that it can handle domain-specific process variation more effectively than conventional variant analysis methods.