Enhancing the explainability of the default probability model using the logic learning machine: A comparison between native “white boxes” machine learning techniques

This study suggests the implementation of the Logic Learning Machine (LLM) methodology to model the default probabilities of a dataset of American firms. This advanced supervised Machine Learning technique, which has the advantage of being inherently “white box”, was developed using a lean low-code development platform that allows the use of the paradigm of visual block programming. The probability default model for an optimal credit risk management was solved using both a statistical regression and a classification approach. The performance obtained in our case study was then compared with that of a Classification and Regression Tree (CART), one of the few supervised Machine Learning techniques that can be considered natively “white box”. The results achieved by the LLM proved to be superior both in terms of performance and explainability compared to those obtained with the CART.