Computational experiment: Comparison of traditional and nuclear clustering methods

This article presents the results of solving various classification problems by the algorithms proposed in this study. Computational experiments were carried out in such a way that their results were comparable with already known solutions. This made it possible to assess the strengths and weaknesses of the minimax approach to data analysis and determine the ways for further research. Of the traditional methods of cluster analysis, linguistic data analysis was chosen, since the method of nuclear clustering has "grown" and is developing in this approach. The two main problems of data processing - accuracy and speed - with large dimensions of processed arrays can be solved by developing accurate and fast algorithms and using new info communication technologies on which these algorithms are implemented. The present work is devoted to solving these two problems. A nuclear clustering method is proposed within the framework of structural or data mining (Data Mining) and the possibilities of its operation in a distributed environment are evaluated. Indeed, the emergence of modern technologies for structural data analysis Data Mining is associated with a new round in the development of data processing tools and methods. The goal of this approach is to uncover hidden rules an patterns in datasets, since the human brain cannot identify more than two or three relationships, even in small samples. Mathematical statistics often turns out to be effective only for testing hypotheses already obtained by Data Mining, and the formulation of hypotheses is performed automatically in the course of a computational experiment. Today this technology is widely used in business (banking, insurance, trade), telecommunications. Foreign experience shows that the use of Data Mining can give an economic effect (10-70 times) as high as in comparison with the initial costs.

Ҳали таржима қилинмаган