Missing data characteristics and the choice of imputation technique: an empirical study

Abstract

One important characteristic of good data is completeness. Missing data is a major problem in the classification of medical datasets. It leads to incorrect classification of patients, which is dangerous to health management of patients. Many imputation techniques have been employed to solve this problem, but these techniques are without recourse to the characteristics that cause the missingness. In this paper, we investigated the causes of missing data in a medical dataset and proposed multiple imputation technique to solving the problem of missing data. A 5-fold-iteration multiple imputation was employed. The whole missing values in the dataset was regenerated 100%. The imputed datasets were validated using extreme learning machine (ELM) classifier. The results show improvement on the accuracy of the imputed datasets. The work can, however, be extended to compare the accuracy of the imputed datasets with different classifiers.