Process discovery on deviant traces and other stranger things

The modelling of business processes is an important task to support decision-making in complex industrial and corporate domains. Recent years have seen the birth of the BPM! (BPM!) research area, focused on the analysis and control of process execution quality, and in particular, the rise in popularity of process mining [van12], which encompasses a set of techniques to extract valuable information from event logs. Process discovery is one of the most investigated process mining techniques. It deals with the automatic learning of a process model from a given set of logged traces, each one representing the digital footprint of the execution of a case. Process discovery algorithms are usually classified into two categories according to the language they employ to represent the output model: procedural and declarative. Procedural techniques envisage the process model as a synthetic description of all possible sequences of actions that the process accepts from an initial to an ending state. Declarative discovery algorithms--which represent the context of this work--return the model as a set of constraints equipped with a declarative, logic-based semantics, and that must be fulfilled by the traces at hand. Both approaches have their strengths and weaknesses depending on the characteristics of the considered process.