Locating and Visualizing the Most Valuable Business Processes States in Large Control Policy Graphs Produced by POMDP

This paper is motivated by the problem of identifying and presenting the most valuable states delivered by a partially observable Markov decision process (POMDP) policy graph with large number of edges and nodes. A control policy graph is the result of executing a POMDP algorithm to a welldefined set of system state transitions, and is composed of a prediction about the value achieved by each state transition that occurs in a business process execution environment. In addition to our problem, the business process state transitions are not always controlled (because of workaround occurrence) and are not fully observable (not know exactly the system state). Moreover, along with the increasing complexity of business processes, the control policy graphs produced by POMDP are exploding in number of edges and nodes, making the best solution identification and visualization a difficult task. Therefore, this paper presents a decision making approach dealing with the non observable states, and filtering of large control policy graphs. Summing up, this process offers a support tool to engineer the decision-making, locating the business process execution states that maximize the delivered value for the organization. This work is located in the Enterprise Engineering field and the approach to locate and visualize the most valuable states delivered by a POMDP uses a previous work that explains how to integrate the business processes modelling with the Markov theories.