Khaustova Y. Fuzzy clustering methods based on kernel functions in data mining tasks

The clustering system based on the evolving general regression neural network and self-organizing map of T.Kohonen, is proposed in the thesis. An on-line neuro-fuzzy system for solving data stream fuzzy clustering task and its self-learning procedures based on T. Kohonen's rule are proposed in the thesis. During a learning procedure in on-line mode, the proposed system tunes both its parameters and its architecture. For tuning of membership functions parameters of neuro-fuzzy system the method based on competitive learning is proposed. In the thesis soft probabilistic clustering algorithm of multidimensional data sets that are sequentially fed to processing in on-line mode is investigated. The proposed system solves the tasks of Data Stream Mining when classes are overlapped.