Improving alignment computation using model-based preprocessing

Alignments are a fundamental approach in conformance checking to provide an explicit relation between traces of events observed in an event log and execution sequences of process models. They are robust against intricacies in process models such as duplicate labels and invisible transitions, but at the same time computing them is a time consuming task. In this paper, we argue that precomputed rules may be leveraged to improve on the time needed to compute alignments. To this end, we utilize both structural and behavioral properties of process models to derive rules and we compare events against these rules. A violation in one of the rules indicates a problem in the event. Before alignments are computed, we mark the problematic events as so-called splitpoints. We evaluated this approach on real-life logs as well as benchmarking logs, and the results show that the proposed approach is faster than existing alignment approaches.