一种通过动态聚类训练椭圆形模糊分类器的方法

Approach for Training Fuzzy Classifier with Ellipsoidal Region by Dynamic Clustering

研究了椭圆形模糊分类器的训练问题 .首先从分类器的结构入手 ,分析异类训练样本形成的椭圆相互重叠时 ,两种原因引起样本被误分 ,由此提出两种动态聚类的方案来增加新的分类规则 ,并且采用基于实数编码的遗传算法训练椭圆 .将提出的训练方法应用于实例 ,并与其它训练方法比较 。

Presented an approach to handle the training of fuzzy classifier with ellipsoidal regions. Described structure of fuzzy classifier and analyzes the possible reasons causing misclassification when ellipsoids of different classes overlap. Proposed two approaches based on dynamic clustering to generate a new fuzzy rule, and real-coded genetic algorithm was used to train the ellipsoidal regions. Experiments on two databases showed that the approach is more efficient, especially when ellipsoidal regions overlap heavily.