南京理工大学学报（自然科学版）2013，Vol.37Issue(1)：19-24,31,7.
代价敏感Boosting算法研究
Cost-sensitive boosting algorithms
摘要
Abstract
In terms of the problem of cost-sensitive learning, this paper investigates cost-sensitive extension of boosting. A cost-sensitive boosting learning framework is proposed based on cost-sensitive sampling. Through introducing cost-sensitive sampling in each round of naive boosting, the expectation of cost-sensitive loss is minimized. Under the above framework, two new cost-sensitive boosting algorithms are deduced. Meanwhile, issues of the instability existing in early cost-sensitive boosting algorithms are revealed and explained. Experimental results on UCI ( University of California, Irvine ) data set and CBCL( Center for Biological & Computational Learning) face data set demonstrate: in terms of the cost-sensitive classification problem, cost-sensitive sampling boosting algorithms outperform naive boosting and existing cost-sensitive boosting algorithms.关键词
boosting/代价敏感boosting/代价敏感学习/代价敏感采样Key words
boosting/cost-sensitive boosting/cost-sensitive learning/cost-sensitive sampling分类
信息技术与安全科学引用本文复制引用
李秋洁,茅耀斌,叶曙光,王执铨..代价敏感Boosting算法研究[J].南京理工大学学报（自然科学版）,2013,37(1):19-24,31,7.基金项目
国家自然科学基金(60974129,70931002) （60974129,70931002）
国家科技重大专项(2011ZX04002-051) （2011ZX04002-051）
中央高校基本科研业务费专项资金资助项目(NUST2011YBZM119) （NUST2011YBZM119）
