Discrete modeling and simulation of business processes using event logs

An approach to business process modelling for short term KPI prediction, based on event logs and values of environment variables, is proposed. Ready-for-simulation process model is built semiautomatically, expert only inputs desired environment variables, which are used as features during the learning process. Process workflow is extracted as a Petri Net model using a combination of process mining algorithms. Dependencies between features and process variables are formalized using decision and regression trees techniques. Experiments were conducted to predict KPIs of real companies.