Assessing operational risks in modern financial and credit institutions

Abstract

The article considers modern methods for assessing the criticality of operational risks arising from technological system failures, human errors, and organizational shortcomings. Traditional approaches, such as regression and clustering, have proven inadequate for analyzing the nonlinear and volatile nature of operational risks in the context of digital transformation. To address these challenges, an innovative approach is proposed, leveraging neural network ensembles and explainable artificial intelligence (XAI) technologies. This approach enhances the accuracy and interpretability of criticality risk forecasts. The article presents the results of previous studies in which multi-layer perceptrons (DNNs) and radial basis function networks (RBFNs) were tested for managing operational risks in credit institutions. These models demonstrated high accuracy in assessing the criticality of risks associated with human errors, Informatic Technology (IT) failures, and business process disruptions. DNNs proved effective in analyzing complex data interrelationships, while RBFNs showed high performance in classifying IT incidents. Based on these results, the further development of models using neural network ensembles is proposed to improve forecast accuracy and resilience to new data. XAI methods, such as LIME and Grad-CAM, are applied to interpret model outcomes, ensuring transparency and trust in decision-making processes. The article also outlines directions for future research and practical steps for implementing the proposed approaches in operational risk management systems.