Label-independent feature engineering-based clustering in Public Administration Event Logs

Process mining algorithms infer business models by analyzing Log files derived from the execution of business activities in organizations. In this paper, a label-independent clustering methodology is proposed. It allows an analysis completely agnostic with respect the nature and domain knowledge of the process Logs. The methodology is totally data-driven and it is based on features that do not depend on activity labels and do not need model extraction at all, thus not requiring the four quality dimensions of a mining discovery algorithm to be satisfied. Due to its independence from asset labels, the methodology is very flexible and applicable in different scenarios. The methodology was tested on the process logs of a municipality of twenty thousand inhabitants showing good performances when evaluated using a mining discovering algorithm.