Abstract
Background: Prescriptive Process Monitoring systems aim at recommending, during the execution of a business process, interventions that, if followed, prevent poor performance of the process. Such interventions have to be (i) reliable: they have to guarantee the achievement of the desired outcome or performance and (ii) flexible: they cannot overturn the normal process execution.
Problem: Most of the Prescriptive Process Monitoring solutions perform well in terms of recommendation reliability but provide the users with recommendations expressed in terms of specific activities that have to be executed without caring about their feasibility.
Method: We propose a new Outcome-Oriented Prescriptive Process Monitoring system recommending temporal relations among activities that have to be guaranteed during the process execution. The proposed system is based on a Machine Learning model that learns the correlations between temporal relations among activities and the (positive) outcome of the process. Then, given the prefix of an ongoing process, the model is queried to return the most promising recommendations.
Contribution: The main contribution is that the proposed system softens the mandatory execution of an activity at a given point in time, thus leaving more freedom to the user in deciding the interventions to put in place. This is achieved by providing recommendations that are expressed as Linear Temporal Logic formulas over activities.
Results: The proposed system has been widely assessed using a pool of 22 real-life datasets. The results demonstrate the reliability of the provided recommendations by achieving an F1 score higher than 90% on 18 datasets out of 22.