Resumen
The subject matter of the study is data clustering based on the ensemble of neural networks. The goal of the work is to create a new approach to solving the tasks of clustering in data streams when information is fed observation-by-observation in online mode. The following tasks were solved in the article: the model of neural network ensembles for data clustering was created, the methods of data clustering to process mass data were developed, the methods of online data clustering of data using neural network ensembles working in the parallel mode were developed. The following results were obtained: the operation principles of the ensembles of the Kohonen neural network were formulated and practical requirements for dealing with mass data were specified. The probable approaches to solving these problems were indicated. The operation principle of the ensemble of parallel tuned Kohonen clustering networks was studied. The procedures based on the WTA and WTM principles were used to train layers of the neural network ensemble. Radial basis functions were used to increase the dimension of the input space. The mathematical model was developed for solving the problem of data clustering in online mode. The mathematical model was developed to determine the quality of clustering using the Davies-Bouldin index, which was rewritten for online mode. Conclusions. The paper proposes a new approach to solving the problem of clustering data streams when information is fed observation-by-observation in online mode, provided that the number and shape of clusters are unknown in advance. The main idea of this approach is based on the ensemble of neural networks, which consists of Kohonen self-organizing maps. All members of the ensemble process information that is sequentially fed into the system in parallel mode. Experimental results confirmed the fact that the considered system can be used to solve a wide range of Data Stream Mining tasks.