TITLE:
Why Can Multiple Imputations and How (MICE) Algorithm Work?
AUTHORS:
Abdullah Z. Alruhaymi, Charles J. Kim
KEYWORDS:
Multiple Imputations, Imputations, Algorithms, MICE Algorithm
JOURNAL NAME:
Open Journal of Statistics,
Vol.11 No.5,
October
14,
2021
ABSTRACT: Multiple
imputations compensate for missing data and produce multiple datasets by
regression model and are considered the solver of the old problem of univariate
imputation. The univariate imputes data only from a specific column where the
data cell was missing. Multivariate imputation works simultaneously, with all
variables in all columns, whether missing or observed. It has emerged as a
principal method of solving missing data problems. All incomplete datasets
analyzed before Multiple Imputation by Chained Equations (MICE) presented were misdiagnosed; results
obtained were invalid and should not be countable to yield reasonable
conclusions. This article will highlight why multiple imputations and how the
MICE work with a particular focus on the cyber-security dataset. Removing
missing data in any dataset and replacing
it is imperative in analyzing the data and creating prediction models.
Therefore, a good imputation technique should recover the missingness,
which involves extracting the good features. However, the widely used
univariate imputation method does not impute missingness reasonably if the values
are too large and may thus lead to bias. Therefore, we aim to propose an
alternative imputation method that is efficient and removes potential bias
after removing the missingness.