Modelling Data-Aware Stochastic Processes - Discovery and Conformance Checking

Process mining aims to analyse business process behaviour
by discovering process models such as Petri nets from process executions recorded as sequential traces in event logs. Such discovered Petri
nets capture the process behaviour observed in a log but do not provide insights on the likelihood of behaviour: the stochastic perspective.
A stochastic Petri net extends a Petri net to explicitly encode the occurrence probabilities of transitions. However, in a real-life processes, the
probability of a trace may depend on data variables: e.g., a higher
requested loan amount will trigger additional checks. Such dependencies are not described by current stochastic Petri nets and corresponding stochastic process mining techniques. We extend stochastic Petri
nets with data-dependent transition weights and provide a technique for
learning them from event logs. We discuss how to evaluate the quality
of these discovered models by deriving a stochastic data-aware conformance checking technique. The implementations are available in ProM,
and we show on real-life event logs that the discovery technique is competitive with existing stochastic process discovery approaches, and that
new types of stochastic data-based insights can be derived.