MaRz: A Fast, Transparent Fuzzy Machine Learning Technique

There is significant interest in fast machine learning and in explainability. This paper’s contribution is a novel, but straightforward, fuzzy model that learns on the fly, is accurate, and explains its conclusions in a literal manner. MaRz (Machine Learning in Realtime with Fuzziness) treats each record as fuzzy and applies classical fuzzy center-of-gravity calculations. In the interest of trustworthiness, MaRz does not attempt a generalized form of explanation. Instead, it shows the specific data that most contributed to the output and allows those data to be tested in the context of the remaining data. It places at the user’s discretion how many such data to provide and thereby increase the explanation. The contribution of this paper is to demonstrate a machine learning approach for categorization and regression of competitive accuracy that is, at the same time, novel, real time, and explainable.