Concept drift detection and localization in process mining
Rafael Gaspar de Sousa, Sarajane Marques Peres, Marcelo Fantinato, Hajo A. Reijers
Abstract
Business processes are subject to changes over time due to the need for adaptation and flexibility to a complex environment. Detecting drift as soon as possible and identifying the process elements involved, lead to a much better understanding of the process behavior, which can be a competitive edge for businesses. However, most existing approaches focus on each of these two tasks separately. Isolated approaches do not always have interfaces between them that allow you to combine solutions effectively for each corresponding task. In such cases, using the two isolated solutions together is neither feasible nor even useful from the point of view of a business analyst. This paper proposes an integrated approach to detect and locate concept drifts based on an online setting for trace clustering. Experiments with synthetic event logs with different types of control-flow changes showed that concept drifts can be detected and located efficiently.