Using process mining to model interarrival times: investigating the sensitivity of the ARPRA framework

Abstract

Accurately modeling the interarrival times (IAT) is important when constructing a business process simulation model given its influence on process performance metrics such as the average flow time. To this end, the use of real data from information systems is highly relevant as it becomes more readily available. This paper considers event logs, a particular type of file containing process execution information, as a data source. To retrieve an IAT input model from event logs, the recently developed ARPRA framework is used, which is the first algorithm that explicitly integrates the notion of queues. This paper investigates ARPRA's sensitivity to the initial parameter set estimate and the size of the original event log. Experimental results show that (i) ARPRA is fairly robust for the specification of the initial parameter estimate and (ii) ARPRA's output represents reality more closely for larger event logs than for smaller logs.