Abstract. The problem of fuzzy clustering of multivariate observations is considered and a group of Kohonen neural network adaptive self-learning algorithms is proposed. The algorithms allow on-line possibilistic fuzzy clustering with variable fuzziness level and are characterized with computational simplicity and great flexibility when operating under conditions of a priori uncertainty about the nature of data distribution in clusters.

Keywords: fuzzy clustering, fuzzifier, Kohonen neural network, self-learning algorithm.