A new method to deal with missing data using Taguchi's T-method

Abstract

Taguchi’s T-method is one of the Mahalanobis-Taguchi (MT) system proposed by Gen-ichi Taguchi. It is used for various forecasting purposes. The T-method is often compared to multiple regression analysis because the data type applied is the same in both methods. One advantage of the T-method over multiple regression analysis is its ability to deal with missing data. Unlike multiple regression analysis, the T-method can be analyzed without removing missing samples. In this study, we propose an improved method that further improves the accuracy of missing data while maintaining the features of the T-method. Specifically, we first improve the T-, Ta-, and Tb-methods using bootstrapping for data with missing values and then investigate a new method for dealing with missing data. Next, by calculating the prediction accuracy using simulations under various models, we examine whether there is a trend toward superiority or inferiority in multiple regression analysis, the T-, Ta-, and Tb-methods as well as in the improved methods over the T-, Ta-, and Tb-methods in the case of missing data. The results show that the improved Tb-method is more accurate than the conventional one, regardless of missing mechanisms. The improved T- and Ta-methods are more accurate in some cases.