摘要
针对决策树泛化能力差,容易产生过拟合问题,提出基于随机化属性选择和决策树组合分类器。首先运用随机化邻域属性约减产生多个分类较高的属性子集;其次每个属性子集作为分类回归树(CART)的输入,训练多个基分类器;最后对得到的多个分类精度结果进行投票融合的方式获得最后的分类结果。实验表明,提出的随机属性选择和决策树集成算法有效性。
Aiming at the problems of poor generalization ability and easy to overfitting in decision tree.We introduce an ensemble classifier based on randomized attribute selection and combination of decision tree. Firstly,multiple subsets with higher accuracy are produced by use of randomization adjacent attribute reduction; Secondly each attribute subset as classification and regression tree's input; Finally the final classification result is obtained by multiple classification results fusion. The experiment result shows that the proposed random attribute selection and decision tree integration algorithm is effective.
出处
《贵州师范大学学报(自然科学版)》
CAS
2016年第1期98-102,共5页
Journal of Guizhou Normal University:Natural Sciences
关键词
过拟合
随机化
决策树
分类器
融合
overfitting
randomization
decision tree
classifier
fusion