Rule Summarization Method for Rule-Based Machine Learning

Abstract

Explainable AI, which can explain the basis of AI decisions, has been attracting attention in recent years. One type of explainable AI is rule-based machine learning. One advantage of this method is that the model itself is a white-box type that is represented by a large number of rules. In actual business settings, this method can lower the hurdle to adoption because the prediction trends can be known in advance. However, the number of rules in the model is increasing due to the complexity of recent data, and it is difficult to check each rule. In this paper, we propose a method to present a summary of rules by grouping similar rules and visualizing the relationships among the groups.