The paper applies the most common methods of Explainable Artificial Intelligence to a model based on the random forest approach that has shown a good ability to predict defaults of Italian firms. The analysis aims at identifying the most relevant variables for the model and evaluates their contribution to the estimation of the probability of default.
The use of Explainable Artificial Intelligence methods allows us to clarify the logic underlying the decision-making process of a model based on the random forest approach, showing how it is able to exploit a large set of predictive variables and, compared with traditional statistical models, to give greater importance to indicators that have a non-linear relationship with firm default, such as indicators of liquidity.
