Abstract
Process mining techniques can be used to discover process models from event data. Often the resulting models are complex due to the variability of the underlying process. Therefore, we aim at discovering declarative process models that can deal with such variability. However, for real-life event logs involving dozens of activities and hundreds or thousands of cases, there are often many potential constraints resulting in cluttered diagrams. Therefore, we propose various techniques to prune these models and remove constraints that are not interesting or implied by other constraints. Moreover, we show that domain knowledge (e.g., a reference model or grouping of activities) can be used to guide the discovery approach. The approach has been implemented in the process mining tool ProM and evaluated using an event log from a large Dutch hospital. Even in such highly variable environments, our approach can discover understandable declarative models.