Title: Reinforcement learning for next best action recommendation in process data
Authors: Yoann Valero - Universite de Technologie de Troyes (France) [presenting]
Leonard Arnold Ebongue Ebaha - Universite de Technologie de Troyes (France)
Frederic Bertrand - IRMA/Universite de technologie de Troyes (France)
Myriam Maumy - IRMA/Universite of Technology of Troyes (France)
Abstract: Predictive business process monitoring (PBPM) aims at predicting the future of running process instances, be it for the next event or the remaining sequence of events (suffix). An event is characterized at least by a triplet made of a unit identifier, an activity the unit can go through, and the time of activity execution. The aim is to set some stepping stones for reinforcement learning in PBPM. Indeed, When making predictions for processes, there is no guarantee that these predictions will be advantageous. Thus, we have developed a method allowing for the next best action recommendation using an agent-based reinforcement learning method. This method has proven to be viable, even with low amounts of data points. In addition, this algorithm allows the recommendation of the entire remaining activities and avoids sub-optimal paths for ongoing units. We tested this algorithm on both public and real-life business process data successfully.