Fig. 5

Distribution of NRMSE values for different imputation approaches for 10% total missing. A–D correspond to changing γ using the COPDGene Data Set 1 p = 300 n = 100. Each figure contains box plots of NRMSE (across 50 simulation repetitions) for the following approaches: 100% Accuracy Imputation, Best Possible Imputation, Mechanism-Aware Imputation, MCAR Only algorithm Imputation, and MNAR Only algorithm Imputation. 100% Accuracy Imputation refers to the NRMSE that would have resulted if the random forest classifier achieved 100% accuracy in step one of MAI, and the Best Possible Imputation refers to the NRMSE that would have resulted if the best imputed values from both the MCAR Only algorithm Imputation and the MNAR Only algorithm Imputation are selected