Extension of incremental linear discriminant analysis to online feature extraction under nonstationary environments

Abstract

In this paper, a new approach to an online feature extraction under nonstationary environments is proposed by extending Incremental Linear Discriminant Analysis (ILDA). The extended ILDA not only detect so-called "concept drifts" but also transfer the knowledge on discriminant feature spaces of the past concepts to construct good feature spaces. The performance of the extended ILDA is evaluated for the benchmark datasets including sudden changes and reoccurrence in concepts. © 2012 Springer-Verlag.