基于局部密度和纯度的自适应k近邻算法

Adaptive k Neighbor Algorithm based on Local Density and Purity

关键词:  k的可信度  自适应k值  KNN分类

Key words:  credibility of k  adaptive k  KNN classification

摘要:

[目的]针对K最近邻(K-Nearest Neighbor,KNN)算法中k值的选取通常是人为设定,而且通常是固定的缺点,研究如何更好地选取k值。[方法]引入k的可信度的概念,提出一种基于局部密度和纯度的自适应选取k值的方法,并将其引入到传统的KNN分类算法中。[结果]该算法合理的考虑了样本的局部密度、纯度与选取k值的关系,不仅解决了k值的选取问题,并且避免了固定k值对分类的影响。[结论]该算法是有效的,可以得到较高的准确率,但算法的时效性有待提高。

Abstract:

[Objective] Aiming at the selection of parameter k value(usually fixed) in KNN algorithm is usually set by users,we should study how to better select k values.[Methods] This paper introduces the concept of the credibility of k,and proposes an improved adaptive selection of k values based on the local density and purity,and introduces into the traditional KNN classification algorithm.[Results] The algorithm is reasonable to consider the relationship between the local density and purity and the selection of k values,which not only solves the problems of choosing k values,but also avoids the influence of fixed k value on classification.[Conclusion] The algorithm is effective and can get higher accuracy,and the timeliness is also enhanced.