摘要
已有的以k-最近邻(kNearest Neighbor,kNN)规则为核心的分类算法,如模糊kNN(FuzzykNN,FkNN)和证据kNN(EvidentialkNN,EkNN)等,存在着两个问题:无法区别出样本特征的差异以及忽略了邻居距训练样本类中心距离的不同所带来的影响.为此,本文提出一种模糊-证据kNN算法.首先,利用特征的模糊熵值确定每个特征的权重,基于加权欧氏距离选取k个邻居;然后,利用邻居的信息熵区别对待邻居并结合FkNN在表示信息和EkNN在融合决策方面的优势,采取先模糊化再融合的方法确定待分类样本的类别.本文的方法在UCI标准数据集上进行了测试,结果表明该方法优于已有算法.
The classification algorithms based on k Nearest Neighbor （kNN） rule, such as Fuzzy kNN （FkNN） and Evi- dential kNN （EkNN）, has two problems：the differences of the sample features cannot be recognized and the effect of fuzziness that aroused by the different distances between neighbors and the center of classes is not taken into account. In order to overcome the limitations, the fuzzy-evidential kNN（FEkNN）algorithm is proposed. First, the features＇ weights are determined by the features＇ fuzzy entropy values and k neighbors are selected according to the weighted Euclidean distance. Then samples are classified by the method, which fuzzify memberships of its neighbors first and then fuse the information. And this method combines the advantage of FkNN in information expression with that of EkNN in decision-making. Meanwhile, neighbors are distinguished by their informa- tion entropy values. The presented method is tested on the UCI datasets,and the results show that the proposed method outperforms the other kNN-based classification algorithms.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2012年第12期2390-2395,共6页
Acta Electronica Sinica
基金
国家自然科学基金(No.60974063
No.61175059)
