Towards Explaining Process Variants with Examples

Abstract

Understanding how and why business processes unfold differently across subsets of cases, commonly referred to as process variants, is essential to identifying improvement opportunities. Existing approaches to process variant analysis largely depend on machine learning models that classify cases into variants and then use model explanations to highlight the features driving such classifications. Prior user studies in artificial intelligence have shown that these model explanations can be complemented through the selection of natural examples that better capture these differences. Thus, process variant analysis can be enriched by elevating the explanation from highlighting features that better describe their differences to selecting the cases containing these features. Inspired by prior work in process variant analysis, we propose a framework that extracts, quantifies, and leverages declarative behavioral rules to identify these cases. Particularly, we identify two types: cases that are representative of the process variants (prototypes) alongside cases that shore up their frontiers (boundary cases), as means to complement existing process variant approaches.