Zhernova P. Data streams fuzzy clustering in conditions of unknown number of clusters

In the thesis proposed the ensemble of T. Kohonen’s self-organizing maps, which is based on using the online method of K-means. This approach allows processing the information in online mode that is fed to the input of the system. Unlike existing clustering methods, the use of the ensemble approach allows to bypass the problem when the number of classes is unknown in advance because each of the Kohonen’s networks is configured for its own number of clusters. The method based on the ensemble approach, using the neural self- organizing T. Kohonen’s maps, has been improved, which allowed using additional hidden layer of the neural network to increase the dimension of the input space. A neuro-fuzzy ensemble of T. Kohonen’s self-organizing maps has been developed for data stream clustering, when the use of the improved C-means and additional neural layer method is able to process information that is linear inseparable, as well as to process clusters of arbitrary form. This allows processing data of high dimensionality and avoiding the concentration of norms effect. The ensemble of self-organizing maps has been improved for the clustering of high- dimensional data streams, which processes information entering the system using several approaches: probabilistic and possibilistic.