融合改进K近邻和随机森林的机器学习方法

摘要

A fused machine learning method was proposed by improving K-nearest neighbor (KNN) and random forest learning methods.Nearest neighbor classification was executed by calculating the distance between the features for classification and the center of each training classes,to enhance the robustness and improve the efficiency of K-nearest neighbor method.The multiple output of improved KNN classifier was transferred to binary output through random division,and it was used to build decision functions of each decision node in random forest,to reduce the error-decision rate of the data reached at decision nodes,in the end to improve the object classification rate of random forest learning method.Experimental results show that the classification rate for handwriting numerical objects classification of the proposed method is higher than the classical learning methods including Knearest neighbor,Adaboost,support vector machine and random forest.%对K近邻和随机森林学习方法进行改进,提出一种融合的机器学习方法.通过计算待分类特征与训练库中各个类中心之间的距离,进行最近邻分类,增强K近邻学习方法的鲁棒性,提高其运算效率;通过随机划分将改进KNN分类器的多元输出转化为二元输出,用其构建随机森林中各个决策结点的决策函数,降低各决策结点数据的错分率,提高随机森林学习方法的目标分类正确率.实验结果表明,该方法对手写数字目标的分类正确率高于经典的K近邻、Adaboost、支持向量机和随机森林学习方法.