Pattern Based Sequence Classification

Abstract

The text highlights the challenges of analyzing massive amounts of data, particularly in the context of business processes using process mining. Outliers or irregular behavior in the data can negatively impact processing and clutter process models, leading to less useful paths. The objective is to automatically extract process models from the data, and an automated method for removing irregular behavior from event logs is introduced. This method significantly improves the quality of identified process models and scales well to large datasets. Since the effectiveness of filtering strategies depends on the event log, users can interactively filter activities and directly view the filtered process model from the event log using a slider-based approach. Ultimately, the choice of included activities is left to the user. The method is tested using actual occurrence log collections from enterprise process oversight and hospital environments. The results demonstrate that the newly proposed activity filtering approaches yield process models that are more behaviorally specific compared to conventional frequency-based filtering methods