Construction of an active learning algorithm that takes surrogate model features into account by using decision tree analysis

Abstract

In generally, the creation of surrogate models presents the challenge of preparing a large amount of training data. Regarding this problem, Integration Neural Network (INN) has been proposed that can make highly accurate predictions even with a small amount of training data. Therefore, I devised an active learning method that takes into account the characteristics of INN so that we can efficiently prepare training data when creating surrogate models. Hence, I constructed an algorithm for active learning by combining an initial learning data generation method that designs experiments to capture the behavior of the entire area to be learned and a method for selecting additional learning points based on decision tree analysis. In consequence, the algorithm met the requirements of the case study and demonstrated its usefulness for efficient data preparation.