摘要
分类器选择是一种设计多分类器系统的有效方法,从给定候选分类器集中挑选出一个子集,使得该子集集成性能最佳。现有的分类器选择方法大多采用基于集成精度的随机搜索方法,但巨大的搜索复杂度限制了它们在更大系统中的应用。该文提出一种新的选择标准——IWCECR及一种基于IWCECR的启发式搜索算法,在手写体数字识别的实验中,从20个候选分类器中挑选子集,结果表明,该方法具有较高的搜索效率,在子集集成性能方面仅次于穷举法。
Classifier selection is an effective way to design multiple classifier systems. The goal of c：assifier selection is to select a subset of classifiers from a given set of candidate classifiers, to achieve the best combination performance. At present, most of classifier selection methods use the stochastic search based on combination accuracy. The burden of complexity of such search limits their practical applicability for larger systems. This paper presents a new selection criterion—— Improved Weighted Count of Errors and Correct Results（IWCECR） and a new heuristic search method based on IWCECR. In experiments of handwritten digit recognition, it selects a subset from 20 candidate classifiers. Results show that search efficiency of the method outperforms others and in respect of combination performance, the method also has high efficiency, although is lower than exhaust algorithm.
出处
《计算机工程》
CAS
CSCD
北大核心
2008年第2期206-208,共3页
Computer Engineering
基金
国家自然科学基金资助项目(60675006
60475003)
北京科技大学基金资助项目
关键词
分类器选择
搜索算法
选择标准
手写体数字识别
classifier selection
search algorithm
selection criterion
handwritten digital recognition
