训练集容量对决策树分类错误率的影响研究

Proportion of Train-Set's Influence on Error-Rate of Decision Tree

数据挖掘算法必须在实际数据集上进行验证,而数据集容量是有限的,训练集比例过低会导致训练不足,训练集比例过高会导致算法评价过于乐观。针对训练集容量对评价效果的影响问题,对25个UCI数据集的不同比例训练集运用决策树算法C4.5,得出不同训练集容量对决策树分类错误率的影响关系。实验结果表明,训练集比例至少为50%时才能使分类错误率达到相对平稳。

Algorithm in Data Mining must be validated upon real dataset,but the amount of sample in any dataset is limited.Excessively low proportion of train-set will cause inadequate training,and excessively high proportion of train-set will cause optimistic evaluation.For proportion of train-set's influence on evaluation,C4.5 is used upon different proportion of train-sets from 25 UCI dataset,Proportion of Train-Set's Influence on Error-Rate of Decision Tree describing is found.Results show that proportion of train-set needs to be at least 50% for a comparatively stable error-rate.