TABLE 7.

Training and inference duration for each imputation method in seconds. We use the wall-clock run time to measure the durations for training, including hyperparameter optimization and inference for all datasets with MCAR missingness pattern and all fractions shown in Table 6. Because training and inference durations depend heavily on the dataset size, we first calculate the durations’ mean and relative standard deviation for each imputation method on every dataset. Second, we average those mean durations and relative standard deviations for the imputation methods and present them as Mean duration and Rel. SD separately for Training and Inference.

Imputation method	Training		Inference
	Mean duration	Relative standard deviation	Mean duration	Relative standard deviation
Mean/mode	0.005	0.550	0.029	0.171
k-NN	41.204	0.254	7.018	0.602
Random forest	226.077	0.119	24.048	0.236
Discriminative DL	6,275.019	0.405	440.389	0.211
VAE	71.095	0.099	11.215	0.085
GAIN	878.058	0.312	137.966	0.083