SELF-EVOLVING NEURAL NETWORKS FOR RULE-BASED DATA-PROCESSING

Two training algorithms for self-evolving neural networks are discusse d for rule-based data analysis, Efficient classification is achieved w ith a fewer number of automatically added clusters, and application da ta is analyzed by interpreting the trained neural network as a fuzzy r ule-based system. The learning vector quantization algorithm has been modified, acquiring the self-evolvement character in the prototype neu ron layer based on sub-Bayesian decision making, The number of require d prototypes representing fuzzy rules is automatically determined by t he application data set, This method, compared with others, shows bett er classification results for data sets with high noise or overlapping classification boundaries. The classifying radial basis function netw orks are generalized into multiple shape basis function networks, The learning algorithm discussed is capable of adding nea neurons represen ting self-evolving clusters of different shapes and sizes dynamically, This shows a clear reduction in number of neurons or the number of fu zzy rules generated, and the classification accuracy is increased sign ificantly, This improvement is highly relevant in developing neural ne tworks that are functionally equivalent to fuzzy classifiers since the transparency is strongly related to the compactness of the system.