决策树算法中过度拟合的研究与解决方案  

Research and Resolve Over-fitting of Decision Tree Algorithm

作　　者：王琴竹[1] 

机构地区：[1]运城学院公共计算机教学部,山西运城044000

出　　处：《运城学院学报》2011年第2期53-54,57,共3页Journal of Yuncheng University

基　　金：山西省高校科技开发项目(20091151);运城学院2009年度院级基础研究项目(JC-2009017)

摘　　要：决策树算法广泛应用于模式识别和机器学习等领域,用来解决与分类相关的问题。决策树算法中的过度拟合会在很大程度上影响到最终的分类结果。针对过度拟合产生的原因,采用悲观错误剪枝方法,对学生成绩决策数据进行分析,得出影响学生成绩的重要因素。实验表明,该方法可以得到尽可能短的分类规则,有效地提高了决策树的性能。Decision tree algorithms are widely used in the field of pattern recognition and machine learning,and used to solve problems related with the classification.The result of classification will be largely affected by over-fitting problem in the decision tree algorithm.According to the reasons of over-fitting,this paper analysis student achivement decision data by pessimistic error pruning method and get some important factors that affect students＇ performance.Experiments show that the method can get a classification rules as short as possible,effectively improve the performance of the decision tree.

关 键 词：决策树 过度拟合 剪枝 悲观错误剪枝法 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]