基于赤池信息准则的分类回归决策树剪枝算法  

Pruning algorithm of classificaiton and regression decision tree based on Akaike information criteria

作　　者：杨贵军[1] 孟杰[1] 王双喜[2] 

机构地区：[1]天津财经大学中国经济统计研究中心,天津300222 [2]中国农业银行股份有限公司郑州二七支行,郑州450000

出　　处：《计算机应用》2014年第A02期147-150,共4页journal of Computer Applications

基　　金：国家社科基金资助重大项目(09&ZD040);国家自然科学基金资助面上项目(11471239);天津财经大学研究生科研基金资助项目(2014TCB02)

摘　　要：决策树剪枝可以提高决策树的分类准确度。代价复杂度剪枝(CCP)等常用的剪枝算法,都以降低决策树的误判率作为剪枝依据。引入赤池信息准则(AIC)评价决策树的优良性,并提出了基于AIC的决策树剪枝算法,将分类正确概率和复杂度的综合评价作为剪枝依据。通过实例分析,基于AIC的剪枝算法能够得到高分类准确度的决策树,并没有出现过拟合或剪枝不充足等问题。Decision tree pruning can improve the prediction accuracy of decision tree. Cost-Complexity Pruning（ CCP）and other commonly used pruning algorithms take reducing misclassification rate as the pruning basis. Akaike Information Criterion（ AIC） can be used as criterion to assess decision tree. A new decision tree pruning algorithm based on AIC was proposed, and the comprehensive evaluation of the probability of correct classification rate and the complexity of decision tree was used as the pruning basis. In the applied example, the pruning algorithm based on AIC made the decision tree classify more accurately without overfitting and insufficient pruning.

关 键 词：赤池信息准则 代价复杂度剪枝 LOGISTIC回归 决策树剪枝 

分 类 号：TP311.11[自动化与计算机技术—计算机软件与理论]