由最大同类球提取模糊分类规则

Extracting Fuzzy Rules from the Maximum Ball Containing the Homogeneous Data

为提高模糊分类规则的有效性和可解释性,该文提出一种基于最大同类球的模糊规则提取方法。首先,每个样本根据与最近异类之间的距离确定一个最大同类球。然后根据各个同类球中样本之间的包含关系和独有性对同类球进行约简。再根据约简后的同类球建立MA分类器的模糊规则前件。MA(Mamdani-Assilan)二分类器的模糊规则后件参数学习以加权分类错误平方最小化为目标函数,采用共轭梯度法求解后件参数。KEEL标准数据集中的12个10折交叉验数据集的对比分类实验验证了该方法的有效性。

In order to improve the interpretability and effectiveness of the fuzzy classifier rules, this paper presents a new method to extract the fuzzy rules based on the maximum ball only containing the homogeneous data. At first, every sample constructs a maximum ball in the light of the shortest distance to heterogeneous samples. Then those balls are reduced according to the relation of inclusion and the unique among the samples that the ball encloses. Then the fuzzy rules are constructed with the reserved balls. The parameters learning of the antecedent part of the classifier are based on the minimization of the weight misclassification quadratic error and resolved with the conjugate gradient algorithm. The experiments on 12 benchmark datasets with 10 folds are performed to demonstrate the validity of the classifier.