A probabilistic approach for threaded process learning

We present a formal probabilistic framework for process learning using log files. We divide the process into separate threads each with an activity. An activity is the set of events that occur in the process from beginning to end. Our proposed learning algorithm is proven to be correct by induction. The learned process is then modelled using a workflow-net which is an extension to Petri nets. Our choice is motivated by the fact that Petri nets handle concurrency and goal reachability, or soundness that can be theoretically established. We define a mathematical merge model which turns merge problems expressed as processes into algebric representations. While we do not address the problem of efficiency, we formally demonstrate that this framework guarantees soundness, and goal reachability.