Toward behavioral business process analysis

This work adresses the problem estimating correlations between process observables and KPI/SLA violation metrics. This analysis is aimed at providing business process owners with a suggestion of potential causality between groups of process parameters and violations. Obviously, static regression analysis cannot establish causality. We introduce a notion of Behavioral Analysis of business process instances and argue that, under suitable assumptions, Granger correlation can be used to highlight potential causality between business process observables and violations of leading KPIs. To handle the combinatorial explosion of the process observables search space, we describe a game theoretical approach for identifying (sets of) time-shifted process attributes to be tested.