Abstract
Scheduling environments are usually dynamic and vary with time. It is necessary that the scheduling method is flexible enough for modifications or changes during production, without interrupting actual operations. Recent researches indicate that applying inductive learning technologies is one of the useful ways to solve these kinds of problems. However, when learning from imbalanced data (almost all the examples are labelled as one class while far fewer objects are labelled as the other class), these methods have poor predictive ability to identify minority instances. This is because most inductive learning algorithms assume that maximizing accuracy on a full range of cases is the goal, and this results in very poor performance for cases associated with the low-frequency class. In this study, we introduce a novel knowledge acquisition algorithm called 'granular computing model' for imbalanced data and integrate this method into a scheduler within a simulated flexible manufacturing system (FMS) environment. Compared with costs adjusting, cluster-based sampling techniques and decision tree (C 4.5), the experimental results indicate that our approach can dramatically increase the predictive ability of minority examples while improving classification performances.