Data classification with ensembles of one-class support vector machines and sparse nonnegative matrix factorization

Abstract

The paper presents a method for data classification with ensemble of one-class classifiers based on data segmentation. Each data class is partitioned with the nonnegative matrix factorization (NMF) algorithm with sparse constraints. It allows splitting of the input data into compact and consistent data clusters with automatic determination of a number of clusters. Data partitions are fed to an ensemble composed of a number of one-class support vector machine (SVM) classifiers. The proposed method shows high accuracy and fast classification.