Abstract: For the purpose of reducing redundancy of ensemble learning，a selective classifier ensemble algorithm based on subgraph strategy is proposed.It trains a set of classifiers and constructs a complete undirected graph with weight by using individual and diversity value between individuals，and chooses individuals with larger diversity based on subgraph strategy to construct ensemble members.By choosing Support Vector Machine（SVM） as basis classifier，experimental study is conducted on several data sets and performance of proposed method is compared with that of common Bagging and Adaboost.Experimental results show that the approach obtains better ensemble accuracy.

Key words: subgraph, diversity, ensemble learning, Support Vector Machine（SVM）

摘要： 为了去除集成学习中的冗余个体，提出了一种基于子图选择个体的分类器集成算法。训练出一批分类器，利用个体以及个体间的差异性构造出一个带权的完全无向图;利用子图方法选择部分差异性大的个体参与集成。通过使用支持向量机作为基学习器，在多个分类数据集上进行了实验研究，并且与常用的集成方法Bagging和Adaboost进行了比较，结果该方法获得了较好的集成效果。

关键词: 子图, 差异性, 集成学习, 支持向量机