缺失森林算法在缺失值填补中的应用

Application of MissForest Algorithm for Imputing Missing Data

目的介绍R环境下缺失森林算法在缺失值填补中的应用并评价其填补效果。方法通过实际数据阐述填补估算流程,比较缺失森林算法与直接删除法处理缺失数据的效果。结果当数据缺失率为10%时,缺失森林算法填补的效果明显优于删除法;当数据缺失率在20%时,两种方法处理缺失值的效果都不太理想,效果相近。当缺失率达50%时,3种类型的变量估算的误差已经较大,两种方法的估算效果均欠佳。结论缺失森林算法在软件操作上简便,并且对数据结构和分布的要求宽松,可充分利用现有记录的信息,能较为准确地反应调查的真实情况,在实际工作中具有较好的应用价值。

Objective To introduce the principle of missForest algorithm and its basic R procedure in imputing missing data,and to assess the imputation effects of missForest. Methods Based on real data sets with missing variables and different missing rate,we introduce R procedure of missForest and compare the imputation results between missForest and deletion method. Results MissForest outperforms deletion method as missing rate is 10%. As the missing rate is increasing by 20%,there are no obvious differences for these methods and the imputation effects of these methods dealing with missing data are unsatisfactory. Whereas missing rate is more than 50%,the relative error of three kinds of variables for these methods is increasing dramatically,neither method is appropriate. Conclusion MissForest is more attractive than other multiple imputation methods for its easy and simple usage in software,moreover it does not require assumptions about the distribution and structure of the data. With this new method,we can make the most use of the data in hand and have more reliable results,so it is worth using widely in practice.