Toward a new generation of log pre-processing methods for process mining

Abstract

Real-life processes are typically less structured and more complex than expected by stakeholders. For this reason, process discovery techniques often deliver models less understandable and useful than expected. In order to address this issue, we propose a method based on statistical inference for pre-processing event logs. We measure the distance between different segments of the event log, computing the probability distribution of observing activities in specific positions. Because segments are generated based on time-domain, business rules or business management system properties, we get a characterisation of these segments in terms of both business and process aspects. We demonstrate the applicability of this approach by developing a case study with real-life event logs and showing that our method is offering interesting properties in term of computational complexity.