关键词: 计算机应用, 集成分类, 特征子空间, DNA微阵列数据, 支持向量机

Abstract:

Low classification accuracy is one of the major restrictions to the use of microarray data in clinical diagnosis. To solve this problem, by taking consideration of the characteristics of high-dimension and small sample of DNA microarray dataset, a novel classification approach was proposed, which is named feature subspace ensemble classification. In this approach, first, hierarchical clustering and signal-noise ratio feature selection strategy were used to project the training dataset from original space to a non-redundant feature gene space. Then, mass feature subspaces were extracted to construct training subsets, and support vector machine was used to train base classifiers. Finally, the majority voting strategy was applied to classify the testing samples. The effectiveness and feasibility of the proposed approach was verified on four benchmark microarray datasets. Experiment results show that better classification performance can be obtained by the proposed approach than majority of the current ensemble classification approaches on microarray data.

Key words: computer application, ensemble classification, feature subspace, DNA microarray data, support vector machine