Abstract: The ensemble classifiers train its base classifiers using the datasets which are generated by some rules.This paper presents a fuzzy membership function based on kernel method to divide the training set into two parts，one is easy to classify while another is hard.Two different base classifiers are trained for fitting them；those two kinds of classifiers are integrated as base classifiers.This method is applied to classify the UCI benchmark datasets，and the experimental results show that this method is superior to Bagging and AdaBoost algorithms on the higher performance.

Key words: fuzzy memberships, kernel function, ensemble classifiers, dataset partition

摘要： 组合分类器通过在输入空间中依据一定的规则生成数据集来训练成员分类器。提出一种新的基于核函数的模糊隶属度方法用来分隔数据集，并依据数据集中样本的模糊隶属度将它们分为相对难分和相对易分的数据子集，根据两个数据子集的难易程度训练不同的分类器。并用得到的两类分类器作为成员分类器生成组合分类器。将该组合分类器应用到UCI的标准数据集，实验表明该方法比Bagging和AdaBoost算法具有更好的性能。

关键词: 模糊隶属度, 核函数, 组合分类器, 数据集分割