Enhancing Machine Learning Interpretability Through a Graphical User Interface

Alexandre K. Tabouret, Eric Michael Vernon, Naoki Masuyama, Yusuke Nojima

Abstract: In designing machine learning algorithms, there is a balance to be struck between interpretability and performance. Typically, simpler algorithms that are easier for humans to understand tend to have lower classification performance compared to more complex and less transparent ones. Over the years, several methods have been proposed to minimize this trade-off as much as possible. This paper builds on a previous study that introduced a partially explainable classifier using white and black box models. It presents a simple graphic user interface (GUI) specifically designed for human users to better understand the mentioned classifier. This paper provides an overview and a few experimental results acquired using the GUI.