一种交叉验证和距离加权方法改进的KNN算法研究  

A nimproved KNN algorithm based on cross validation and distance weighting

作　　者：黄光华 殷锋[1] 冯九林 HUANG Guang-hua;YIN Feng;FENG Jiu-lin(School of Computer Science and Technology,Southwest Minzu University,Chengdu 610065,China)

机构地区：[1]西南民族大学计算机科学与技术学院,四川成都610041

出　　处：《西南民族大学学报（自然科学版）》2020年第2期172-177,共6页Journal of Southwest Minzu University(Natural Science Edition)

基　　金：国家自然科学基金项目(61105061);四川省科技支撑项目(14ZC1767);国家社会科学基金项目(16ZDA166);西南民族大学研究生创新型科研项目(CX2019SZ62)。

摘　　要：传统的KNN算法在大样本的情况下对试验样本有较强的一致性结果.但该算法每一次都是在样本全局的基础上进行归类计算,这在很大程度上消耗了空间,增大了算法空间复杂度;且算法在样本分类不平衡时,预测偏差相对偏高.针对这些问题,提出了基于交叉验证和距离加权的改进KNN算法(Weighted cross-validation KNN,简称WCKNN),经过实验验证,本算法有效减小了算法的空间复杂度;同时,也获得了更好的分类性能.The traditional KNN algorithm has strong consistency results for test samples in the case of large samples.However,the algorithm performs classification calculation on the basis of the sample globally,which consumes space to a large extent and increases the space complexity of the algorithm.Moreover,when the sample classification is unbalanced,the prediction bias is relatively high.Aiming at these problems,a modified KNN algorithm based on cross-validation and distance weighting(WCKNN)is proposed.The experimental results show that the algorithm effectively reduces the space complexity of the algo-rithm.At the same time,it also gets better classification performance.

关 键 词：交叉验证 距离加权 KNN算法 大样本 归类问题 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]