Comparison of G-Means Algorithms and Kohonen Network in Solving Clustering Problems

Abstract

Purpose: In this paper, the question of how to improve a self-organizing neural network consisting of a bundle of clustering algorithm and a multilayer perceptron for data verification tasks in the absence of training pairs is considered. Design/methodology/approach: The most popular clustering algorithm is the Kohonen network, but today it is not the only algorithm capable of performing the task quickly and accurately. The paper compares the Kohonen network and the G-Means algorithm. The principle of operation of these two algorithms is briefly analyzed. The accuracy of these algorithms and the speed of their learning are compared. Findings: By conducting experiments, conclusions were drawn about the speed and accuracy of the algorithms. Originality/value: The relevance of this work lies in the fact that the preparation of training pairs for intelligent systems and the process of learning with a teacher is a resource-intensive task. The systems of self-learning algorithms under consideration will significantly increase the learning rate, as well as eliminate the need for manual classification of data and the creation of training pairs for the perceptron, which in turn will allow you to create a self-learning system with the ability to generalize and predict.