I have a dataset with 18000 rows and 192 columns. I have a specific column with more than 2000 rows missing.
I've tried some types of imputations but they just take to long or just seem not good enough. I wonder if it's okay to do some kind feature selection of the dataset to get the most important variables before doing a multivariate missing value imputation (like MICE). This way I could work with maybe 5 to 15 relevant variables for the imputation and then place the imputation on the original dataset.
Is there any problems to this aproach? like bias in the final dataset.
Thanks in advance