Comparison Method for Handling Missing Data in Clinical Studies

Abstract

Missing data is an issue that cannot be avoided. Most data mining algorithms cannot work with data that consist of missing values. Complete case analysis, single imputation, multiple imputations, and kNN imputation are some methods that can be used to handle the missing data. Each method has is own advantages and disadvantages. This paper compares of these methods using datasets in clinical studies, chronic kidney disease, Indian Pima diabetes, thyroid, and hepatitis. The accuracy of each method was compared using several classifiers. The experimental results show that kNN imputation method provides better accuracy than other methods.