Abstract
In the case of regulated or highly standardized processes, event log-based Process Mining frequently leads to less meaningful homogeneous process flows in process models, providing less to no valuable insights for in-depth process analytics. This concealment of process complexities overlooks underlying variations and hinders further process optimization. Machine learning for log enrichment (ML4LE) allows for analysis at a finer granularity with reduced manual effort. To achieve this, ML4LE leverages unsupervised Machine Learning techniques to infer activity-level subgroups and enrich event logs. By using the contextual information underlying the raw data in databases, this method detects subgroups and integrates them into the process model through activity label splitting to derive subvariants. This approach not only facilitates the creation of more detailed and meaningful process models but also avoids unnecessary overcomplexity. An empirical computational study that compares the proposed method to the classical non-preprocessing approaches demonstrates its efficacy and practical significance, reinforcing its potential in real-world applications.
Fabian Gwinner
PHD Candidate
My research interests include Machine Learning, especially Deep Learning, XAI, Decision Support Systems