局部一致性的信息熵Relief特征加权算法

摘要

In order to improve the adaptability and robustness of traditional Relief algorithm, by integrating margin maximization,information entropy and the local consistency of the classification,the margin maximizing objective function was structured.Then by applying optimization theory,some of useful theoretical results were derived.Based on this,a new Relief feature weighting algorithm which was named as LIE-Relief-T(Local consistency information entropy Relief algorithm based two-class data)based on two-class data was proposed.LIE-Relief-T algorithm was extended to feature weighting algorithm for multi-class data which was named as LIE-Relief-M(Local consistency information entropy Relief algorithm based multi-class data).Experimental results using UCI and gene expression datasets showed that the new Relief feature weighted algorithm had lower classification error rate and better adaptability and robustness to noise and field points.%为了改善传统Relief算法适应性和鲁棒性差的缺陷,融合间距最大化、信息熵和分类局部一致性,构造了新的间距最大化目标函数,并进一步对目标函数进行优化,得到一些新的理论结果.在此基础上提出了新的基于两类数据的Relief特征加权算法LIE-Relief-T(Local consistency information entropy Relief algorithm based two-class data),并将其扩展到多类数据的特征加权算法LIE-Relief-MLocal consistency information entropy Relief algo-rithm based multi-class data).利用UCI和基因表达数据集进行实验验证,结果表明该新的Relief特征加权算法分类错误率较低,对噪声和野点表现出了更好的适应性和鲁棒性.