Subgroup Discovery in Process Mining Mohammadreza Fani Sani 1 , Wil van der Aalst 1 , Alfredo Bolt 1 and Javier Garc´ ıa-Algarra 2 1 Eindhoven University of Technology, Eindhoven, The Netherlands {M.Fani.Sani,w.m.p.v.d.aalst,a.bolt}@tue.nl, 2 Telefonica, Madrid, Spain fco.javier.garciaalgarra@telefonica.com Abstract. Process mining enables multiple types of process analysis based on event data. In many scenarios, there are interesting subsets of cases that have deviations or that are delayed. Identifying such subsets and comparing process mining results is a key step in any process mining project. We aim to find the statistically most interesting patterns of a subset of cases. These subsets can be created by process mining algorithms features (e.g., conformance checking diagnostics) and serve as input for other process mining techniques. We apply subgroup discovery in the process mining domain to generate actionable insights like patterns in deviating cases. Our approach is supported by the ProM framework. For evaluation, an experiment has been conducted using event data from a large Spanish telecommunications company. The results indicate that using subgroup discovery, we could extract interesting insights that could only be found by spitting the event data in the right manner. Key words: Process Mining · Subgroup Discovery · Pattern Mining · Performance Management · Quality of Metrics. 1 Introduction Our society, organizations and IT systems depend on processes. Products and services can only be delivered efficiently and effectively when processes are run- ning as planned. Process mining aims to discover, monitor, and enhance processes by extracting knowledge from event data that can be extracted from almost all modern [1]. Process Mining is able to bridge the gap between Business Process Modeling (BPM) and data driven methods like data mining and machine learning [2]. Process mining is able to analyze the actual processes without relying on sim- plistic models. There are basically two main types of data-driven analysis [3]: – Predictive analysis: involving techniques that extract knowledge and rules to predict or classify samples, such as classification, regression and time series algorithms. – Descriptive analysis: involving techniques that discover interesting knowl- edge about samples and their attributes to explain the data (e.g. association rules).