EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

4319. Leveraging Manifold Learning for Constrained Counterfactual Explanations in Process Outcome Prediction

Invited abstract in session MC-27: XAI in Business Processes, stream Mathematical Optimization for XAI.

Monday, 12:30-14:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Jari Peeperkorn
Research Center for Information Systems Engineering (LIRIS), KU Leuven
2. Alexander Stevens
KULeuven

Abstract

Harnessing the potential of machine and deep learning architectures within Predictive Process Analytics (PPA) has yielded promising results, particularly in predicting future process outcomes. However, the opacity inherent in these algorithms poses a significant challenge for human decision-makers, limiting their understanding of the rationale behind the predictive outputs. In response, counterfactual explanations, serving as human-understandable ‘what if’ scenarios, offer intuitive insights into the decision-making process behind undesirable predictions. Nevertheless, generating counterfactual explanations encounters specific challenges when dealing with the sequential nature of business process cases in PPA. To address this, our paper introduces REVISED+, a novel data-driven approach aimed at generating more feasible and plausible counterfactual explanations tailored to the complexities of sequential process data. The counterfactual generation algorithm is guided through high-density regions of the process data distribution, and Declare language templates are used to capture sequential patterns among the process activities. This enables REVISED+ to improve the feasibility and plausibility of the generated counterfactual explanations. Finally, we assess the validity of our counterfactuals with metrics that define the key properties of a good counterfactual.

Keywords

Status: accepted

Back to the list of papers