Missing values is an actual yet challenging issue confronted
in researches in machine learning and data mining. Missing
values may generate bias and negatively affect the quality
of the supervised learning process or the performance of
data clustering algorithm [20], [21]. In fact, general analysis
methods, such as principle component analysis [9], multivariate
supervised classification or clustering analysis, require a
complete data matrix. Dealing with missing values is to find
an approach with the ability to fill in these previously unknown
whilst maintaining (or approximate as closely as possible) the
original distribution of the data under examination. Imputation
is a term that denotes a procedure of replacing the missing
values in a dataset by some plausible values. Missing data are
often replaced by zeros or less often, by an average expression
over the row, or ’row average’ [1]. These statistical approaches
are rarely optimal, since they do not take into consideration
the correlation structure represented within the data.