Performance Evaluation for Class Center-Based Missing Data Imputation Algorithm

Abstract

The imputation method should be able to reproduce the actual values in the data or Predictive Accuracy (PAC) and maintaining the distribution of these values or Distributional Accuracy (DAC). However, in most studies, evaluation of imputation performance was measured based on classification accuracy. On classification issues, class center-based methods for missing data imputation are developed and outperform other methods for numeric and mixed data types. This paper will be evaluated the accuracy of class center-based methods for missing data imputation, which has been modified by considering the correlation between attributes. A class center-based method for missing data imputation produces an average value of r is 0.96, with the lowest average value for MSE and DKS is 0.04 and 0.03. This result shows that the imputation method is more efficient and can maintain the actual data value distribution.