【摘要】 为提高决策树分类算法的精度,通过比较几种经典的决策树分类算法,提出了基于样本选取的改进的决策树分类算法.改进算法基于决策树精度与样本的相关性较大以及决策树只能得到局部最优解的事实,通过反复迭代寻找较优样本,从而在不改变决策树分类算法的前提下,得到较好的决策树分类算法.该算法不针对某个决策树,只利用输入和输出的反馈信息进行迭代,因此通用性较好.实验证明,该改进算法与ID3,C4.5算法平均错误率的比值约为0.82∶1.22∶0.92.更多还原

【Abstract】 To raise the accuracy of decision tree classification algorithms,an improved decision tree classification algorithm based on samples selection was proposed by comparing several classical decision tree classification algorithms.This improved algorithm searches better samples through a constantly iterative process based on the facts that the correlation between decision trees’ accuracy and samples is large and decision trees can only get a local optimal solution.As a result,a better decision tree classification algorithm can be obtained under the condition of not changing the decision tree classification algorithm.The improved algorithm is not aiming at a decision tree and it carries through iteration only based on some feedback information of input and output,so its universality is better.Experimental results show that the ratio of the average error rates of the improved algorithm and the ID3,C4.5 algorithms is about 0.82 to 1.22 to 0.92.更多还原