Classification of datasets with imputed missing values: does imputation quality matter?