A class center based approach for missing value imputation

Missing value imputation (MVI) is the major solution method for dealing with incomplete dataset problems in which the missing attribute values are replaced from a chosen set of observed data using some statistical methods, such as mean/mode, machine learning, or support vector machine methods. Although machine learning MVI approaches may produce reasonably good imputation results, they usually require larger imputation times than statistical approaches. In this paper, a Class Center based Missing Value Imputation (CCMVI) approach is introduced for producing effective imputation results more efficiently. It is based on measuring the class center of each class and then the distances between it and the other observed data are used to define a threshold for the later imputation. The experimental results based on numerical, categorical, and mixed data types of datasets show that the proposed CCMVI approach outperforms the other MVI approaches for both numerical and mixed datasets. In addition, it requires much less imputation time than the machine learning MVI methods.