基于模糊积分的纠错输出解码研究

针对基于纠错输出编码的多类分类中解码策略的选择问题,提出一种基于模糊积分的解码方法.将每个基分类器看作一个证据源,其概率输出提供了测试样本所属类别的可能性;同时基于基分类器信任度和混淆矩阵对模糊密度的赋值进行了改进;最后求取最大的模糊积分值所对应的类别为分类输出.基于UCI数据集和经典解码方法的比较实验验证了本文所提方法作为二符号解码策略的可行性和有效性.

Decoding based on fuzzy integral in error-correcting output codes for multi-classification

The decoding strategy is a hot issue in the multi-classification based on error-correcting output codes. As to the problem, a new decoding strategy based on fuzzy integral is proposed in this paper. In this method, each base classifier is taken as an evidence source and provides the probability of samples belonging to each meta-class. Meanwhile, the fuzzy density is modified based on confusion matrix and classifier confidence to estimate the classification ability of different classes. Finally, the class label with the maximal fuzzy integral value is the predicted output. The experiments based on UCI dataset validate the feasibility and efficiency of the presented approach.