Interpretability of neural networks: a credit card default model example

Recently developed techniques aimed at answering interpretability issues in neural networks are tested and applied to a retail banking case

CLICK HERE TO VIEW THE PDF

Ksenia Ponomareva and Simone Caenazzo show the feasibility of overcoming the interpretability hurdles around the application of neural networks in the estimation of credit risk for a portfolio of credit cards

Historically, the widespread use of advanced deep learning models in sensitive fields such as medicine and finance has been hindered by a fundamental lack of interpretability regarding the outcomes of such models. Simpler techniques such as linear or logistic