Project R-5462

Contributing to the evaluation framework for process discovery: The importance of the process discovery purpose and a novel generalization measure. (Research)

Process mining allows the discovery of a control-flow model from observed process behavior, captured in event logs. The last 15 years, many discovery algorithms have been developed, each with their own strengths. To support further scientific progress in this domain, the community is in need of a strong methodological evaluation framework. Several authors have already contributed to this framework, but several important issues remain. Two of these issues will be tackled within this proposal. Firstly, I will connect the purpose of process discovery to the appropriate evaluation criteria and formalize this in the purpose-perspectives framework. This framework will guide the practitioner to determine the right balance between the different evaluation perspectives, and will help academia agree upon a set of evaluation criteria for model comparison. Secondly, I will develop a new evaluation measure which calculates the probability that the observed event log could have been generated by the discovered model. This measure will allow practitioners and academia to assess if the discovered model does not contain too much unobserved behavior. This will be the first measure to answer the important question if a discovered model has become too general.

01 October 2014 - 30 September 2015