Recommendation of Process Discovery Algorithms: a Classification Problem

Process mining techniques extract knowledge from event logs of information systems. Process discovery is a process mining category, focused on discovering process models. The applicability and effectiveness of process discovery algorithms depend on event log’s features. Selecting the right algorithms is a tough task due to the variety of variables involved and the complexity of obtaining logs features. To choose a suitable discovery algorithm the traditional approaches use empirical assessment. The metrics to perform this assessment are not applicable to all algorithms. Besides, empirical evaluation is time consuming and computationally expensive. The present paper proposes a new approach that, based on event log characteristics, recommends the discovery algorithms to be used. A new technique of sub-processes diagnosis is proposed for characteristics extraction. The recommendation procedure is formalized as a typical classification problem. This approach could be useful for large event logs and unclear processes analysis.