Naučno-tehničeskij Vestnik Informacionnyh Tehnologij, Mehaniki i Optiki (Feb 2024)
Solving the problem of preliminary partitioning of heterogeneous data into classes in conditions of limited volume
Abstract
In the context of the formation of heterogeneous data that differ significantly in nature, even of a small volume, it becomes necessary to analyze them for decision-making. This is typical for many high-tech industrial fields of human activity. The problem can be solved by bringing heterogeneous data to a single view and then dividing it into clusters. Instead of searching for a solution for each data element, it is proposed to use the division of the entire set of normalized data into clusters, and thereby simplify the process of isolating the cluster and making a decision on it. The essence of the proposed solution is the automatic grouping of objects with similar data into clusters. This allows you to reduce the amount of analyzed information by combining a lot of data and perform mathematical operations already for the cluster. When splitting, it is proposed to use the theory of fuzzy logic. The possibility of such an approach is due to the fact that different objects always have several characteristics by which they can be combined. These signs are often not obvious and are poorly formalized. A hierarchical modification of the AFC fuzzy clustering method based on the operation (maxmin) of the fuzzy similarity ratio is proposed. The basic concepts and definitions of the proposed method of automatic partitioning of a set of input data, a step-by-step scheme of the corresponding cluster procedure are considered. The efficiency of the proposed method is demonstrated by the example of solving the problem of forming a traffic flow. A numerical experiment has shown that the developed algorithm allows you to automatically analyze heterogeneous data and stably divide them into classes. The application of the proposed modification allows for the preliminary partitioning of data into clusters and allows reducing the volume of analyzed data in the future. There is no need to consider the objects in each case separately.
Keywords