| Summary: | 碩士 === 逢甲大學 === 資訊工程所 === 91 === This thesis proposes an evolutionary approach to designing the accurate classifier with a compact fuzzy-rule base using an intelligent genetic algorithm IGA and a grid partition of feature space. To design an accurate fuzzy classification system, the flexibility of membership functions is increased by using a parameterized trapezoidal membership function. Since the number of possible fuzzy rules is exponentially increased with the numbers of input features, it is an intractable task to obtain compact classifiers for high-dimensional classification problems, especially when the number of parameters in a membership function is increased. The design of fuzzy classifiers is formulated as a large parameter optimization problem with three objectives: 1) high classification ability, 2) small number of fuzzy rules, and 3) small total number of antecedent conditions. IGA hybrids the advantages of conventional genetic algorithms and orthogonal experimental design and can efficiently solve large parameter optimization problems. IGA and an efficient chromosome encoding are used to effectively solve the investigated problem, in which the membership function and fuzzy rule are simultaneously determined. Extensive computer simulations demonstrate that the proposed method is capable of efficiently solving classification problems to generate accurate and compact fuzzy classifiers with fuzzy rules of high interpretability.
|
|---|
