基于均衡7×2交叉验证的模型选择方法  

Model Selection by Balanced 7×2 Crossvalidation

作　　者：杜伟杰[1] 王瑞波[2] 李济洪[2] 

机构地区：[1]山西大学数学科学学院,山西太原030006 [2]山西大学计算中心,山西太原030006

出　　处：《太原师范学院学报（自然科学版）》2013年第1期27-31,共5页Journal of Taiyuan Normal University:Natural Science Edition

摘　　要：交叉验证策略广泛应用于分类问题的模型比较和模型选择中.文章提出一种均衡7×2交叉验证并给出了相应的构造方法.文章以分类回归树(CART)为考察模型,对比了采用均衡7×2交叉验证、组块3×2交叉验证、标准5折和10折交叉验证在模型选择中的性能.模拟结果表明,在小规模数据集上,均衡7×2交叉验证方法选择到真模型的概率明显高于其余三种交叉验证的选择方法.Cross-validation is a widespread strategy for classification model comparison and selection. We introduced a balanced 7 X 2 cross-validation （CV） strategy,and further provided a corresponding construction method. We compared the performance of balanced 7 X 2 CV, blocked 3X 2 CV, standard 5-fold and 10-fold CV for CART model selection. The simulation results showed that,on small data set, the probability of selecting the true model by balanced 7 X 2 cross- validation was obviously higher than the other three cross-validation strategies.

关 键 词：模型选择 交叉验证 均衡7×2验证交叉验证 均衡设计 

分 类 号：O213.9[理学—概率论与数理统计] TP391[理学—数学]