Classification of Data Chunks Using Proximal Vector Machines and Singular Value Decomposition

Abstract

Data production grows at an unprecedented increasing rate in every research and technical field. Furthermore, with the explosion of sensors networks and proprietary/legacy classifiers, like those used by banks for assessing the credit approvals, the data production and modeling is done locally, where only the local classifiers are available. In order to find a global classification rule, the ensemble classification paradigm proposes several methods of aggregation. In this paper, starting from a set of classifiers obtained by using a recently developed classification technique, known as Regularized Generalized Eigenvalues Classifier, we present a novel way of aggregating linear classification models using the Singular Value Decomposition. Using artificial datasets, we compare the developed algorithm with a voting scheme, showing that the proposed strategy allows a reduction in computational cost with a classification accuracy that well compares with the original method.