动态加权投票的多分类器聚合

Multiple Classifier Fusion with Dynamic Weighted Voting

在使用多分类器系统时,一种流行的方法是采用简单的多数投票策略来聚合多分类器。然而,当各个独立的分类器的性能不统一时,这种简单的多数投票规则会对分类结果造成负面影响。引入一种新的动态加权函数来聚合多个分类器,动态加权函数通过增加分类结果距离样本最近的分类器的权值来提高分类器的性能。在UCI机器学习数据库中的几个现实问题数据集上的实验结果表明动态加权的多分类器聚合方法比简单的多数投票方法能取得更好的分类结果。

When a multiple classifiers system is used, one of the most popular methods to realize the classifier fusion is the simple majority voting. When the performance of each single classifier is not consistency, the efficiency of this simple majority voting generally results affected negatively. Introduces a new function of dynamic weighting for classifier fusion. This new dynamic weighting procedure is to reward the individual classifier with the nearest neighbor to the input pattern. Experimental results on the several real-problem data sets from the UCI machine learning database repository show that dynamic weighting strategies is better than the simple majority voting scheme.