Continuously evolving classification using time-varying AR modeling

Abstract

Continuously evolving classification is an important problem in pattern recognition applications. This paper deals with continuously evolving classification of signals subjected to an abrupt, change. This can be considered as a two-category classification problem: before an instant N/sub r/ the signal under study belongs to one class, after N/sub r/, it belongs to another one. The aim of our study is to understand the continuously evolving classification behavior when applied to this kind of signals. In this paper, a time-varying autoregressive modeling using Walsh functions (TVARW) is presented and the model parameters are used to classify signals subjected to abrupt changes. This model is compared with other classical autoregressive ones. It is shown that this modeling gives better classifying results than the other ones.