Geração genética de classificador fuzzy intervalar do tipo-2

• Autor(a) principal: Pimenta, Adinovam Henriques de Macedo
• Data de Publicação: 2009
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/444
• topic: Inteligência artificial; Sistemas Fuzzy; Algoritmos genéticos; Sistemas Fuzzy-genético; Algoritmos genéticos auto-adaptativos; Sistemas fuzzy genéticos; Geração automática de regras fuzzy; Interval type-2 fuzzy set systems; Genetic algorithms; Self- adaptative genetic algorithms; Genetic fuzzy systems; Automatic generation of fuzzy-rules; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

The objective of this work is to study, expand and evaluate the use of interval type-2 fuzzy sets in the knowledge representation for fuzzy inference systems, specifically for fuzzy classifiers, as well as its automatic generation form data sets, by means of genetic algorithms. This work investigates the use of such sets focussing the issue of balance between the cost addition in representation and the gains in interpretability and accuracy, both deriving from the representation and processing complexity of interval type-2 fuzzy sets. With this intent, an evolutionary model composed of three stages was proposed and implemented. In the first stage the rule base is generated, in the second stage the data base is optimized and finally, the number of rules of the rule base obtained is optimized in the third stage. The model developed was evaluated using several benchmark data sets and the results obtained were compared with two other fuzzy classifiers, being one of them generated by the same model using type-1 fuzzy sets and the other one generated by the Wang&Mendel method. Statistical methods usually applied for comparisons in similar contexts demonstrated a significant improvement in the classification rates of the intervalar type-2 fuzzy set classifier generated by the proposed model, with relation to the other methods.