Performance Comparison of Imputation Methods Using Machine Learning Techniques for Ordinal Missing Data

Objectives: When missing values occur, complete case analysis can cause biased results. In this paper, we discuss imputation methods using machine learning techniques when occurred in ordinal variables. Methods: We consider two techniques, the decision tree and random forest, for of values. use treating variables as ordinal, forest nominal. addition, apply cumulative logistic model. The results are compared with empirical bias, mean squared error accuracy. same applied Korea National Health Nutrition Examination Survey. Results: five categories, yield better performance than case. shows lower bias while higher 3 produces all respects. study, also identified if analysis. Random best performance, parametric method similar to tree. Conclusions: Missing reduce improve performance. If possible, it is recommended impute that reflects meaning order. not treat them at least nominal then impute.