Минимаксный подход к структурному анализу данных методом ядерной кластеризации

Аннотация

Subjects of research: matrixes of proximity, images, directed and undirected graphs Purpose of work: the development and study of minimax model of layered clusterization based on conception of monotonous proximity function Methods of research: the methods of discrete mathematics, linguistic data analysis, classification, cluster analysis, image processing. Software was realized on C++, MATLAB, C#. The results obtained and their novelty: - development of core clusterization parametric model. Based on conception of monotonous proximity function; - the method of image segmentation based on core clusterization, and its’ efficiency in comparing with method of normalized cut and k-mcans; - the procedure of protein sequences multiple alignment based on core clusterization was evaluated in comparing with known software such as CLUSTAL and DIALIGN. Practical value: developed software can be used in different applications in area of data and image processing, particularly in bioinformatics problems solution. Degree of embed and economic effectivity: the results of dissertation were used as software package for image processing and teaching needs in information system department of Murom Institute of Vladimir state University University and as software for computing diagnostics of cholecystitis different forms in Republican Scientific Center of Emergency Medicine, Uzbekistan Ministry of Healthcare. Efficiency is social impact on research, treatment of patients, and education. Field of application: Data Mining and decision-making systems in biology and healthcare.