决策树中避免过度拟合的方法  

Methods to Avoid Overfitting in Decision Trees

作　　者：王黎明[1] 刘华[1] 

机构地区：[1]武汉理工大学计算机科学与技术学院,湖北武汉430070

出　　处：《软件导刊》2006年第10期80-82,共3页Software Guide

摘　　要：通过学习训练数据集来构造分类树的策略可能无法达到最好的泛化性能。随机噪声和某些决策仅基于少量训练数据的情况都会导致决策树的分类精度下降,并且过度拟合训练数据集。避免过度拟合主要是通过对树的剪枝来实现,包括预剪枝和后剪枝。后剪枝方法有很多种,主要从计算复杂性、误差估计和算法理论基础角度分析其中的REP、MEP和规则后剪枝算法。Learning a decision tree through a training set may not lead to the tree with the best generalization performance. The noises in the training set can make the decision tree overfit the training set and reduce the accu-racy of classification. Moreover, the algorithm might be making some decisions toward the leaves based on very little data and may not reflect reliable trends in the training data. Gen- erally, the authors exploit pruning methods to avoid overfitting. There are two methods for pruning, pre-pruning and post-pruning. The paper mainly emphasizes REP,MEP and Rules Post-pruning in term of computational complexity, error estimation and theoretical prin- ciple.

关 键 词：噪声 过度拟合 误差 后剪枝 降低误差剪枝 最小误差剪枝 规则后剪枝 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]