Evolving granular neural network for semi-supervised data stream classification

Abstract

In this paper we introduce an adaptive fuzzy neural network framework for classification of data stream using a partially supervised learning algorithm. The framework consists of an evolving granular neural network capable of processing nonstationary data streams using a one-pass incremental algorithm. The granular neural network evolves fuzzy hyperboxes and uses nullnorm based neurons to classify data. The learning algorithm performs structural and parametric adaptation whenever environment changes are reflected in input data. It needs no prior statistical knowledge about data and classes. Computational experiments show that the fuzzy granular neural network is robust against different types of concept drift, and is able to handle unlabeled examples efficiently.