Business Process Digital Twin Using Process Mining and Deep Learning

A digital twin of a real loan-application process, learned from the BPI Challenge 2012 event log (59,302 work-item events, 8,616 cases). Process mining finds where the time goes: waiting is 99 % of case time. Prediction models estimate the next activity and the remaining time of a running case, on whole cases held out in time. A stochastic simulation in Prosimos answers what-if questions after a validation gate: the mean and the 90th percentile of the cycle time pass, the median fails (-59 %), so what-if numbers are quoted only for the mean, as simulation estimates and not as measured improvements.

Category: Information Science and Technology

Information Science & Technology Project IST05 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.