AN MILP MODEL FOR MULTI-CLASS DATA CLASSIFICATION
Gang Xu, Lazaros G. Papageorgiou · 2006
AbstractThis paper presents a multi-class data classification approach based on hyper-boxes using a mixed integer linear programming (MILP) model. Comparing with other discriminant classifiers, hyper-boxes are adopted to capture the disjoint regions and define the boundaries of each class so as to minimise the total misclassified samples. Non-overlapping constraints are specified to avoid overlapping of boxes that belong to different classes. In order to improve the training and testing accuracy, an iterative solution approach is presented to assign multi-boxes to single class. Finally, the applicability of the proposed approach is demonstrated through two illustrative examples from machine learning databases. According to the computational results, our approach is competitive in terms of prediction accuracy when comparing with various standard classifiers.