Credit risk management is a central issue for the stability and performance of banking institutions. To assess the creditworthiness of their clients, financial institutions have traditionally relied on classical statistical methods, which present several limitations in terms of accuracy and the handling of complex data. The emergence of Data Analytics and Artificial Intelligence (AI) enables the development of more advanced models, capable of capturing nonlinear relationships and accurately anticipating payment defaults. Although the contribution of these techniques to credit scoring is now widely recognized, the systematic comparison of their performance across different empirical contexts and methodological settings remains a key challenge for effective operational adoption. In this perspective, the objective of this article is to rigorously and comparatively evaluate the performance of traditional methods and AI models applied to credit scoring. Specifically, the study compares two classical statistical methods; discriminant analysis and logistic regression with four AI approaches; k-nearest neighbors, Support Vector Machine, Random Forest, and Multilayer perceptron neural networks, using two international banking credit datasets. The models are evaluated using multiple performance metrics to assess their stability and discriminative power across distinct empirical contexts. The experimental results demonstrate the superior performance of AI methods in assessing borrowers’ creditworthiness, confirming their predictive advantage while highlighting the robustness of their performance according to the chosen criteria. These findings suggest that such approaches could be adapted and integrated into Moroccan banks to enhance credit risk prediction and strengthen the management of banking performance. Keywords: Artificial Intelligence; Banking Performance; Prediction; Credit Risk Management; MLP Neural Networks.
Hjouji et al. (Fri,) studied this question.