Constructing knowledge-based artificial neural network with rough sets

摘要

An approach of constructing knowledge-based artificial neural network based on rough sets is proposed. Crude domain knowledge is extracted from the 0-1 table produced from fuzzy information table of example data by a threshold. The extracted initial rules and their accuracy and coverage are used to configure the fuzzy multilayer perceptron-structure and initial weights. An algorithm of attribute-reduction based on information entropy is also proposed in this paper. Results on diagnosises of rice pests show that the performance of this fuzzy neural system is same with that of conventional multi-layer perceptron.