一种适用于不均衡数据集分类的KNN算法

传统的K-最邻近(K Nearest Neighbor,KNN)分类算法在处理不均衡样本数据时,其分类器预测倾向于多数类,少数类分类误差大。针对此问题从数据层的角度改进了传统的KNN算法。先通过K-means聚类算法将少数类样本聚类分组,将每个聚类内的样本作为遗传算法的初始种群;再使用遗传交叉和变异操作获取新样本,并进行有效性验证。最终获取到各类别样本数量基本均衡的训练样本集合。实验结果表明此方法有效改善了KNN算法对少数类分类效果。此法同时适用于其他关注少数类分类精度的不均衡数据集分类问题。

A KNN algorithm for Unbalanced Data Set

Traditional K-Nearest Neighbor(KNN)categorization algorithm,the prediction result of classification was towards the class with more training samples,when it was used to train imbalanced data sets.The classification error of the class with fewer training samples was grave.Aiming at this problem,the traditional KNN algorithm was improved from the data angle: the class with fewer training samples was grouped by using K-means cluster algorithm,the samples in each cluster as the initial population of genetic alg...