中文摘要：

针对基因表达数据高维和小样本的特点,介绍一种基于主成分分析的决策树集成分类算法——旋转森林.首先通过对数据属性集的随机分割,再对子集进行主成分分析变换,保留全部的主成分系数,重新组成一个稀疏矩阵.然后对变换后的数据利用非剪枝决策树集成算法进行分类.再结合ReliefF算法,选用3组基因表达数据验证算法,对比Bagging决策树和随机森林两种集成方法.结果表明旋转森林算法对基因数据具有更好的分类精度,同时验证旋转森林在较低的集成数的情况下,可以取得良好的效果.

英文摘要：

Aiming at the character of high dimensions and small samples of gene expression data, an ensemble classification algorithm by the name of rotation forest based on decision tree was introduced. By splitting the feature set of data, applying the principal component analysis （PCA） on them and then reserving all the coefficients of the principal components, a sparse matrix was rebuilt up. Finally the unpruned decision tree ensemble algorithm was used to classify the transformed data set. Here, combined with the ReliefF algorithm, three groups of gene expression data were choosen to test the rotation forest algorithm, compared with two other ensemble methods： Bagging tree and random forest. The result indicates that the rotation forest has a higher classification accuracy and can get an excellent performance with a low ensemble size all the same.