基于粒计算的K近邻多标签学习算法

K-nearest Neighbor Multi-label Learning Algorithm Based on Granular Computing

K近邻多标签学习算法的近邻点个数取固定值,而没有考虑样本分布的特点,可能会将相似度高的点排除在近邻集外,或者将相似度低的点包含在近邻集内,影响分类器的性能。为此,将粒计算的思想引入近邻集的构建,提出一种新的K近邻多标签学习算法。通过粒度控制,确定近邻点集,使得领域内的样本点有高相似性,且此类样本能进入近邻集。实验结果表明,该算法的大多数评价指标均优于现有的多标签学习算法。

In Multi-label K-nearest Neighbor（ML-KNN） learning algorithm, the number of nearest neighbors is given in prior and its value is chosen without considering the distribution of samples, it is possible that highly similar samples are not in the nearest neighbor or low similar samples are in the nearest neighbor set, which affect the performance of the classifier. In view of this case, a novel ML-KNN algorithm is put forward based on the idea of Granular Computing（GrC）, the nearest neighbor set is constructed with the controlling of the granular hierarchy, and the nearest neighbors of a sample have high similarity and highly similar samples can be added to nearest neighbor set. Experimental results show that most of the evaluation criteria in new algorithm are better than the traditional algorithm.