Learning to act: a Reinforcement Learning approach to recommend the best next activities

  • 2022-06-15 15:29:50
  • Stefano Branchi, Chiara Di Francescomarino, Chiara Ghidini, David Massimo, Francesco Ricci, Massimiliano Ronzani
  • 0

Abstract

The rise of process data availability has recently led to the development ofdata-driven learning approaches. However, most of these approaches restrict theuse of the learned model to predict the future of ongoing process executions.The goal of this paper is moving a step forward and leveraging available datato learning to act, by supporting users with recommendations derived from anoptimal strategy (measure of performance). We take the optimization perspectiveof one process actor and we recommend the best activities to execute next, inresponse to what happens in a complex external environment, where there is nocontrol on exogenous factors. To this aim, we investigate an approach thatlearns, by means of Reinforcement Learning, the optimal policy from theobservation of past executions and recommends the best activities to carry onfor optimizing a Key Performance Indicator of interest. The validity of theapproach is demonstrated on two scenarios taken from real-life data.

 

Quick Read (beta)

loading the full paper ...