Estimation of the Optimal Substitution Value of Intentional-Value-Substitution for Three-Dimensional Problems

Abstract

This paper studies a modeling method that makes prediction error as low as possible under the condition that there is no missing data in a training dataset while some values are missing in the phase of prediction. In our previous research, we proposed Intentional-Value-Substitution learning as well as a method for estimating optimal substitution values for two-dimensional problems. In this paper, we extend the method of estimating the optimal substitution values to the case of three-dimensional problems. It is shown that the substitution values estimated by the proposed method help the models predict outputs for inputs with missing values not only for two-dimensional problems but also for three-dimensional problems.