Predicting makespan in Flexible Job Shop Scheduling Problem using Machine Learning

Abstract

Machine learning tools have experienced a growing interest in the early 2010s, providing efficient predictive approaches for artificial intelligence and statistical analysis. These same prediction methods have also sparked interest in the operations research community for decision-making based on predictive analysis by exploiting massive histories and datasets. This study investigates the potential of machine learning tools to predict the feasibility of a production plan. Production schedules are often not able to adhere to the production plans because production plans are built without accounting for all the detailed requirements that arise at the scheduling level. We show that predicting the feasibility of a production plan with a decision tree yields a precision of around 90% versus 70% in the classical capacity constraints considered in planning tools.