点击次数：

论文类型：期刊论文

发表时间：2015-07-01

发表刊物：ICIC Express Letters

收录刊物：EI、Scopus

卷号：9

期号：8

页面范围：2177-2182

ISSN号：1881803X

摘要：Many real world datasets contain missing values, and partially missing datasets have become a prevailing problem in pattern recognition. In such problems, the rationality of missing value imputation is a key issue. In this paper, missing values are represented as nearest-prototype neighborhoods, which make an efficient use of the prototype information and can be achieved by considering fuzzy cluster covariance matrix. And then, as the nearest-prototype neighborhood representation can constrain the imputation of missing values in rational ranges rather than the entire attribute space, a novel fuzzy c-means algorithm with the neighborhood constrains is proposed, which can avoid the imputations from falling into unnecessary local minima, and enhance the rationality of both the imputation of missing values and clustering results. The proposed algorithm is evaluated with several benchmark datasets and the results demonstrate the better clustering performance of our approach over the compared methods. ? 2015 ICIC International.