Demystifying Noise and Outliers in Event Logs. Review and Future Directions

Various process mining techniques exist, e.g., techniques that automatically discover a descriptive model of the execution of a process, based on event data. Whereas the premise of process mining is clear, i.e., as witnessed by the tremendous growth of the field, data quality issues often hamper the direct applicability of process mining techniques. Several authors have studied data quality issues in process mining, yet, these works primarily propose data pre-processing techniques. An overarching study of the nature of data quality issues, the types of available techniques, and the general possibilities of (semi)-automated outlier/noise detection methods is missing. Therefore, in this paper, we propose a first attempt to structure and study the field of outlier/noise detection in process mining and understand to what degree knowledge on noise and outliers from other domains could advance the process mining field. We do so by answering three central research questions, covering various aspects related to (semi)-automated outlier/noise detection.