Imputation of missing data using ensemble algorithms

Missing data or incomplete data are very common in statistical situations. One way to deal with missing data is to conduct model imputation either one time or multiple times. One of the key problems in analyzing the imputed dataset is to give the valid statistical reference of the parameter estimated, that is, to give a right estimation of the standard error of the interested statistic. This paper proposes the new developed ensemble algorithms as imputation model. In order to realize multiple imputation, we suggest bootstrap sampling the prediction error several times. The properties of the proposed methods are studied by simulation and compared with existing methods. Finally, the methods are applied to analyze one real large dataset, taking the missing mechanism into consideration.