Comparing comparative process mining methods: towards a research agenda

Abstract

Comparing processes is a fundamental analytical task for organizations seeking to understand variation, performance differences, and opportunities for improvement. In domains such as healthcare, such comparisons are particularly relevant, for example when contrasting care pathways of different patient groups, examining how process execution differs over time periods, or benchmarking similar processes across hospitals or care networks. Process mining offers a data-driven foundation to support these types of comparisons by systematically analyzing event data that capture how processes are actually executed. Within the broader field of process mining, comparative process mining has been recognized as a distinct type of analysis, aimed at identifying differences and commonalities between process executions across variants, cohorts, or contexts [1]. Recent studies have highlighted several shortcomings in current comparative process mining methods. Firstly, existing techniques often narrowly focus on syntactic or control-flow aspects, neglecting perspectives such as probabilistic behavior, timing, or resource utilization [7]. Secondly, the definitions of what constitutes a relevant difference remain largely ad hoc, with little theoretical grounding. Finally, visualization techniques frequently fail to accommodate the complexity and multi-dimensionality of comparative findings [6]. The lack of standardization and methodological maturity in the field affects both theoretical progress and applicability. While methods such as log delta analysis enable verbalizing behavioral differences in natural language [3], they remain limited in scope and offer minimal support for multi-perspective or large-scale comparative settings. Likewise, recent methods that incorporate stochastic or statistical perspectives show potential, but they typically focus on localized or aggregate behavioral statistics, offering limited integration with an explicit process model that captures the global structure and relationships between process fragments [7]. To draw up a research agenda for comparative process mining, we propose 1