Study of missing data imputation techniques for the purpose of classification

Abstract

This study evaluates the performances of missing data treatments by comparing the classification results obtained after applying classifiers on contaminated data sets imputed by means of various imputation methods. The objective is to investigate whether a combination of imputation and classification methods could mitigate the potential adverse effect of contamination and missingness in order to achieve satisfactory classification results. For this purpose, simulations have been performed within a well-defined theoretical framework.