摘要
针对传统K最近邻特征投影(KNNFP)算法中假设各维特征对分类的贡献相同而导致分类性能下降的问题,提出一种基于特征加权的KNNFP改进算法(WKNNFP)。改进算法利用ReliefF算法确定特征的权值,使样本的分类效果更好,同时还可以分析各特征对分类的贡献程度,并利用改进算法对轴承故障进行诊断。结果表明,改进算法的诊断率优于传统的KNN和KNNFP算法。
To solve the problem that traditional K nearest neighbor on feature projection assumes that each feature of the samples plays a uniform contribution for classification analysis which lead to lower classification accuracy. A novel feature weighted K nearest neighbor on feature projection is proposed in this paper, in which the reliefF algorithm is used to assign the weights for every feature. By weighting the feature of samples, the better classification results can be achieved. The different contribution to classification performance of every feature can be analyzed. The proposed method is applied to fault diagnosis field which outperforms traditional K nearest neighbors classification methods.
出处
《电子技术应用》
北大核心
2011年第4期113-116,121,共5页
Application of Electronic Technique
基金
教育部社科研究基金青年项目(07JC870006)
教育部社科研究青年项目基金(09YJC870001)
安徽财经大学教研重点项目(ACJYZD200914)
关键词
K最近邻特征投影
特征加权
K最近邻
故障诊断
K nearest neighbor on feature projection
feature weighted
KNN
fault diagnosis