摘要
针对kNN分类算法对不平衡数据进行分类可能偏向多数类的问题,提出了象限壳近邻分类算法。该算法仅选择测试样本象限方向上的最近邻的训练样本来判断其所属类别,从而有效地避免了kNN算法对选取k个最近邻训练样本时可能产生偏向多数类的问题。通过在UCI真实不平衡数据集上的实验,该文提出的分类算法在Recall、F-value和G-mean等评价标准明显优于传统的kNN分类算法。
This paper proposes a Quadrant‐Encapsidated Nearest Neighbor (RBQENN) for short classification method to solve the problem of the conventional kNN (k Nearest Neighbors) method ,i . e .,biasing to majority classes .The proposed algorithm uses the nearest neighbors around the quad‐rant of a testing sample to decide its class label .The experimental results on UCI datasets show that the proposed algorithm outperforms the traditional kNN classification method .
出处
《广西师范学院学报(自然科学版)》
2015年第1期57-62,70,共7页
Journal of Guangxi Teachers Education University(Natural Science Edition)