Kopaliani D. Evolving cascaded neuro-fuzzy system for intelligent data analysis

The thesis is devoted to research and development of evolving cascaded neuro-fuzzy systems for intelligent data analysis in an online mode. The extended neo-fuzzy neuron and its adaptive learning algorithm are proposed. The proposed neuron implements arbitrary order Takagi-Sugeno fuzzy inference and is proven to have enhanced approximating capabilities as well as high operating speed and therefore is an eligible base unit for the sought-for evolving systems. The thesis proposes a number of architectures (both single and multiple output) specifically designed to process nonstationary data in an online mode, utilizing proposed "generalizing" units, that produce the optimal output signal based on signals generated by each neuron in a cascade pool. Such technique has demonstrated its effectiveness in growing data samples especially in case of significant properties drift over time. The evolving system for fuzzy data stream clustering is proposed that is unique in its capability of detecting the optimal number of clusters as it operates. An experimental study of the properties and characteristics of the developed methods is carried out and recommendations on their use in solving practical tasks are proposed.