Comparative Evaluation of Clustering Methods in Working With Big Data

Abstract

The paper considers the problems of using cluster analysis methods in the tasks of processing, analyzing and storing structured and unstructured large-volume data and evaluates the feasibility of their use in various aspects of working with Big Data. The aim of the work is to identify the most preferred of the common data clustering algorithms. To do this, the task was set to conduct a comparative evaluation of the following popular algorithms: hierarchical clustering, k-means, DBSCAN, OPTICS and CURE. The algorithmic complexity of the methods is considered, the stability of algorithms to noise and emissions is analyzed, as well as the potential possibilities of visualizing their results and the scope of economic application are indicated. Conclusions are drawn about the advantages and disadvantages of each presented algorithm when used in the field of Big Data and about the most preferred methods of cluster analysis in various aspects of working with big data.