Dummy training data generation method towards robust estimation of confidence value of semi- automatic agents for multi-class classification

Abstract

In this paper, we propose a dummy training data generation method towards robust estimation confidence value of semi-automatic agents for multi-class classification for the purpose of improving the performance. In the case where machine-learning is used for calculating the confidence value, there is a problem that can calculate accurately only with existing training data. Our approach is using simple random values. As a result of experiments of actual data, we confirmed the improvement of the performance.