A Task Decomposition Algorithm Using Mixtures of Normal Distributions for Classification Problems

Abstract

This paper proposes an algorithm for decomposing a multi-class classification problem into a set of two-class classification problems. The algorithm divides a set of input pattern vectors in each class into subsets according to the distribution of the selected input pattern vectors. The distribution is represented by a mixture of normal distributions, and the number of subsets is defined by using MDL criterion. The algorithm can be applied for constructing an effective modular neural network. We show also the experimental results of the construction and the advantages of the algorithm.