Process remaining time prediction using query catalogs

Abstract

A relevant topic in business process management is the ability to predict the outcome of a process in order to establish a priori recommendations about how to go forward from a certain point in the process. Recommendations are made based on different predictions, like the process remaining time or the process cost. Predicting remaining time is an issue that has been addressed by few authors, whose approaches have limitations inherent to their designs. This article presents a novel approach for predicting process remaining time based on query catalogs that store the information of process events in the form of partial trace tails, which are then used to estimate the remaining time of new executions of the process, ensuring greater accuracy, flexibility and dynamism that the best methods currently available. This was tested in both simulated and real process event logs. The methods defined in this article may be incorporated into recommendation systems to give a better estimation of process remaining time, allowing them to dynamically learn with each new trace passing through the system. © Springer International Publishing Switzerland 2014.