Proper imputation techniques for missing values in data sets

Data mining requires a pre-processing task in which the data are prepared and cleaned for ensuring the quality. Missing value occurs when no data value is stored for a variable in an observation. This has a significant effect on the results especially when it leads to biased parameter estimates. It will not only diminish the quality of the result, but also disqualify for analysis purposes. Hence there are risks associated with missing values in a dataset. Imputation is a technique of replacing missing data with substituted values. This research presents a comparison of imputation techniques such as Mean\Mode, K-Nearest Neighbor, Hot-Deck, Expectation Maximization and C5.0 for missing data. The choice of proper imputation method is based on datatypes, missing data mechanisms, patterns and methods. Datatype can be numerical, categorical or mixed. Missing data mechanism can be missing completely at random, missing at random, or not missing at random. Patterns of missing data can be with respect to cases or attributes. Methods can be a pre-replace or an embedded method. These five imputation techniques are used to impute artificially created missing data from different data sets of varying sizes. The performance of these techniques are compared based on the classification accuracy and the results are presented.