Evaluation and Comparison of Data Imputation Methods on Acxiom Dataset

Abstract

Data imputation is an important step of data analysis. Ad hoc solutions, such as listwise deletion and mean imputation, are often considered unsatisfactory. Aided by the advancement of machine learning research, new approaches have been proposed to estimate missing values using state-of-art statistical models. This article reviews several missing value imputation techniques and compares its performance on Acxiom, a mixed-type, 1600-column consumer demographic and segmentation dataset.