Missing Data Imputation Using a Regime Switching Technique

Abstract

The purpose of this paper is to develop a regime switching technique to optimise mean and regression of a missing data set whose sample is small in size with a low degree of correlation. The data sets were first generated with a simple random method and later treated with the missing completely at random method (MCAR) in order to simulate complete data sets. We classified the data sets with different scenarios of sample size, degree of correlation and percentage of missing data. Moreover, we performed the tests with the missing data imputation techniques, namely: (i) mean imputation (MI), (ii) regression imputation (RI), (iii) regime switching for mean imputation (RsMI), (iv) regime switching for regression imputation (RsRI), (v) average regime switching between mean and regression imputation (aRsMRI), and (vi) weighted regime switching between mean and regression imputation (wRsMRI). The simulation results showed that in the scenario of small sample size and low degree correlation, wRsMRI techniques outperformed other techniques which use MSE evaluate accuracy.