Estimating Activity Start Timestamps in the Presence of Waiting Times via Process Simulation

Process Mining aims to analyze and improve processes to enable organizations to provide better services or products. The starting point of Process Mining is an event log that is extracted from the organization's information systems that support the process' executions. Several techniques require event logs to record the timestamp when process' activities have started and been completed. Unfortunately, information systems do not always record the timestamps when process activities start, preventing the application of these techniques. This paper reports on a technique based on process simulation that aims to estimate the start event timestamps when missing. In a nutshell, the idea is to build an accurate process model from the initial event log without start timestamps, to simulate it with alternative activity-duration profiles, and to select the model with the profile that generates the runs that are the closest to the initial log. This activity-duration profile is used to add the missing, start timestamps to the initial log. Experiments were conducted with two event logs with start timestamps, and aimed at their rediscovery: the results show our estimation of the start event timestamps is more accurate than the state of the art.