Opposite Maps: Vector Quantization Algorithms for Building Reduced-Set SVM and LSSVM Classifiers

Abstract

A novel method, called Opposite Maps, is introduced with the aim of generating reduced sets for efficiently training of support vector machines (SVM) and least squares support vector machines (LSSVM) classifiers. The main idea behind the proposed approach is to train two vector quantization (VQ) algorithms (one for each class, \({\fancyscript{C}}_{-1}\) and \({\fancyscript{C}}_{+1}\), in a binary classification context) and then, for the patterns of one class, say \({\fancyscript{C}}_{-1}\), to find the closest prototypes among those belonging to the VQ algorithm trained with patterns of the other class, say \({\fancyscript{C}}_{+1}\). The subset of patterns mapped to the selected prototypes in both VQ algorithms form the reduced set to be used for training SVM and LSSVM classifiers. Comprehensive computer simulations using synthetic and real-world datasets reveal that the proposed approach is very efficient and independent of the type of VQ algorithm used.