Interpreting business process case outcome prediction with XAI
Track: Full Paper
Abstract: Machine Learning (ML) interpretability techniques were crucial for enhancing Predictive Process Monitoring (PPM) within Business Process Management (BPM), ensuring the strategic integration of ML rather than its mere adoption. VisInter4PPM was proposed previously as a business-oriented approach to visually support interpretability in predictive process monitoring. The VisInter4PPM framework was designed to bridge the gap by providing actionable insights into process predictions. It relied on the results of the SP-LIME interpreter to generate explanations about the influence of each business process activity on the case outcome.
In this paper we present the last version of VisInter4PPM, two results will be presented, the fist one for a synthetic event log, which represented an illustrative health insurance claim management process in a travel agency (a binary class prediction problem); and, a second one for a real event log, which referred to a loan request business process, represented in a real-world event log of a financial institution (a multi-class prediction problem).
The utility of the framework was validated in two design cycles through an expert evaluation. The evaluation confirmed that VisInter4PPM successfully met the needs of business experts by projecting interpretability directly into process models in BPMN notation, which is typically a more familiar working environment for them. This approach not only supported but also enhanced decision-making in complex business environments, making a compelling case for the essential role of ML in modern BPM. This research offered both a methodological framework and empirical evidence essential for advancing ML transparency in BPM, positioning this study as a useful resource for practitioners aiming to navigate and lead in the ML-driven evolution of business processes.
Submission Number: 59
Loading