一种新的基于聚类的多分类器融合算法

提出了一种新的多分类器融合算法，该算法能找出各分类器在特征空间中局部性能较好的区域，并利用具有最优局部性能的分类器的输出作为最终的融合结果。首先，利用各分类器对训练样本进行分类，这样训练样本被划分为正确分类样本和错误分类样本两个集合；接着，对这两个样本集合分别进行聚类分析来划分特征空间，并计算各分类器在特征空间局部区域中的性能；在测试时，选择测试样本周围局部性能最优的分类器的输出作为最终的融合结果。基于ELENA数据集的实验显示了该算法的有效性。

A NOVEL CLUSTERING-BASED MULTIPLE CLASSIFIERS COMBINATION ALGORITHM

An algorithm for combining multiple classifiers is presented, which can find in the feature space the regions where each classifier has best performance. The correctly and incorrectly classified training samples from each classifier are clustered separately to form a partition of the feature space, and the performances of the classifier in each region are calculated. Then, the classifier responsible for the vicinity of the input sample is nominated to label the input pattern. The performance comparison between this algorithm and Kuncheva's CS+DT method, as well as some simple aggregation methods, such as maximum, minimum, average, and majority vote using ELENA data sets, confirm the validity of the proposed combination scheme.