为了解决在分类器集成过程中分类性能要求高和集成过程复杂等问题,分析常规集成方法的优缺点,研究已有的分类器差异性度量方法,提出了筛选差异性尽可能大的分类器作为基分类器而构建的一个层级式分类器集成系统。构建不同的基分类器,选择准确率较高的备选,分析其差异性,选出差异大的分类器作为系统所需基分类器,构成集成系统。通过在UCI数据集上进行的试验,获得了很好的分类识别效果,验证了这种分类集成系统的优越性。
In order to solve the problem of improving the performance of classifying and complexity of combining,the advantages and disadvantages of former combination methods are analyzed,diversity measures are studied,to buildup the combination system used classifiers of greater diversities.First of all,different classifiers are constructed,choosing ones of great precision,then analyzing the diversities,get the classifiers ofmore diversities to formthe combination system.It is a valid combining system method and get excellent result on the experiment upon UCI.