基于精简随机子空间的多生成树集成一类分类算法

由于高维数据通常存在冗余和噪声，在其上直接构造覆盖模型不能充分反映数据的分布信息，导致分类器性能下降.为此提出一种基于精简随机子空间多树集成分类方法.该方法首先生成多个随机子空间，并在每个子空间上构造独立的最小生成树覆盖模型.其次对每个子空间上构造的分类模型进行精简处理，通过一个评估准则(AUC值)，对生成的一类分类器进行精简.最后均值合并融合这些分类器为一个集成分类器.实验结果表明，与其它直接覆盖分类模型和bagging算法相比，多树集成覆盖分类器具有更高的分类正确率.

One-Class Classifier Algorithm Based on Ensemble Multi-Spanning Trees by Pruning Random Subspace Method

Due to the redundancy and the noise in high-dimensional data,a covering model constructed from these data can not reflect their distribution information,which leads to the performance degradation of one-class classifiers. To solve this problem,a pruning random subspace ensemble multi-spanning tree method is proposed. Firstly,several random subspaces are created,and minimum spanning tree covering models are constructed in each subspace respectively. Next,pruning ensembles are applied to each classifier by using an evaluation criterion. Finally,these subspace classifiers are integrated into an ensemble classifier by mean combining. Experimental results show that the proposed covering classifier by ensemble multi-trees has a better correct rate in classification than other direct covering classifiers and bagging algorithm.