Abstract
In this paper, we propose DeepAlign, a novel approach to multi-perspectiveprocess anomaly correction, based on recurrent neural networks andbidirectional beam search. At the core of the DeepAlign algorithm are tworecurrent neural networks trained to predict the next event. One is readingsequences of process executions from left to right, while the other is readingthe sequences from right to left. By combining the predictive capabilities ofboth neural networks, we show that it is possible to calculate sequencealignments, which are used to detect and correct anomalies. DeepAlign utilizesthe case-level and event-level attributes to closely model the decisions withina process. We evaluate the performance of our approach on an elaborate datacorpus of 30 realistic synthetic event logs and compare it to threestate-of-the-art conformance checking methods. DeepAlign produces bettercorrections than the rest of the field reaching an overall accuracy of 98.45%across all datasets, whereas the best comparable state-of-the-art methodreaches 70.19%.