Margin Based Sample Weighting for Stable Feature Selection

摘要：

Stability of feature selection is an important issue in knowledge discovery from high-dimensional data. A key factor affecting the stability of a feature selection algorithm is the sample size of training set. To alleviate the problem of small sample size in high-dimensional data, we propose a novel framework of margin based sample weighting which extensively explores the available samples. Specifically, it exploits the discrepancy among local profiles of feature importance at various samples and weights a sample according to tim outlying degree of its lo cal profile of feature importance. We also develop an efficient algorithm under the framework. Experiments on a set of public microarray datasets demonstrate that the proposed algorithm is effective at improving the stability of state-of-the-art feature selection algorithms, while maintain ing comparable classification accuracy on selected features.

作者: Yue Han Lei Yu

作者单位: State University of New York at Binghamton Binghamton, NY 13902, USA

会议类型: 国际会议

会议名称: 11th International Conference,WAIM 2010(第十一届网络时代管理国际会议)

会议地点: 九寨沟

会议语种:英文

页码: 680-691

在线出版日期: 2010-07-14（万方平台首次上网日期，不代表论文的发表时间）