Self-learning pattern classification using a sequential clustering technique

Abstract

This paper presents an algorithm for the classification of data, based on what we call the self-learning approach. First we mention briefly the fuzzy approach of Ruspini to the problem of pattern classification; from there we set our self-learning approach and we present the problem of classification as that of estimating a partition of the data to be classified. The fundamental tool of this algorithm is the use of numerical filters for estimating a set of parameters which characterize each class. This algorithm has been applied to the recognition of the components of a mixture of normal distributions.