STACKING DOUBLE GENERALIZED NEO-FUZZY NEURON AND ITS FAST HYBRID LEARNING

Abstract

The work is devoted to the development of a hybrid neuro-fuzzy architecture for improving the efficiency of data mining tasks. The subject of the study is the process of approximation of nonlinear dependencies and adaptive learning in data streams conditions using neuro-fuzzy models. The task of the research is to increase prediction accuracy and learning speed by combining stacking principles with generalized neo-fuzzy neurons and a dual signal processing structure. The methods used in this work include a multi-level ensemble or stacking approach, generalized neo-fuzzy neuron models, and a hybrid learning algorithm that integrates gradient-based optimization with recurrent parameter estimation techniques. The result of the study is the proposed Stacking Double Generalized Neo-Fuzzy Neuron (SDGNFN) architecture, which demonstrates improved prediction accuracy, reduced computational complexity and high convergence speed compared to classical neo-fuzzy neurons, conventional neural networks and other ensemble methods. The experimental results confirm the effectiveness of the approach, particularly in online and streaming data scenarios. Further research is focused on extending the proposed model to large-scale real-time systems, improving its robustness under noisy data conditions and exploring its application to a wider range of tasks such as pattern recognition, load forecasting, resource allocation, and control of complex dynamic systems.