Learning Planning Model for Semantic Process Compensation

Abstract

Recent advancements in business process conformance analysis have shown that the detection of non-conformance states can be learned with discovering inconsistencies between process models and their historical execution logs, despite their real behaviour. A key challenge in managing business processes is compensating non-conformance states. The concentration of this work is on the hardest aspect of the challenge, where the process might be structurally conformant, but it does not achieve an effect conform to what is required by design. In this work, we propose learning and planning model to address the compensation of semantically non-conformance states. Our work departs from the integration of two well-known AI paradigms, Machine Learning (ML) and Automated Planning (AP). Learning model is divided into two models to address two planning problems: learning predictive model that provides the planner with the ability to respond to violation points during the execution of the process model, and instance-based learning model that provides the planer with a compensation based on the nearest class when there are no compensations perfectly fit to the violation point.