Study of Fuzzy Classification Systems Using Adaptive and Genetic Algorithms

• Main Authors: Nai Ren Guo, 郭乃仁
• Other Authors: Tzuu-Hseng S. Li
• Format: Others
• Language: en_US
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/2yd675

博士 === 國立成功大學 === 電機工程學系碩博士班 === 93 === 　The dissertation proposes four of fuzzy models to solve the classification problem. First, a hierarchical fuzzy classification model is set up, where several fuzzy subsystems are included. The IF-THEN rules of each subsystem are generated according to the distributions of the feature variables. Two genetic algorithms are utilized to determine the combination of the input features for each subsystem and reduce the number of rules in each fuzzy subsystem, respectively. Secondly, an adaptive fuzzy classification model is established. The confident value of the IF-THEN rule represents the rule weight interpreted as its confident strength. A novel adaptive modification algorithm is developed to tune the confident value of the fuzzy classification model. In the 3rd fuzzy model, the fuzzy feature extraction agent and the fuzzy classification unit are developed by using genetic algorithms to determine the distribution of the fuzzy sets for each original feature variable and the newborn feature variable, respectively. The fuzzy feature extraction agent can validly reduce the original feature dimensions. Finally, an integrated classification model is built, where the genetic algorithm and adaptive grade mechanism are propounded to tune the fuzzy feature extraction agent and fuzzy classification unit, respectively. In order to show the feasibility and efficiencies of the proposed fuzzy classification models, the well-known Wine, Iris and Glass databases are exploited to test the performances.