Innovative Neural Network Approaches to Operational Risk Management in Financial Institutions

Objective: The goal of the study is to create ANN models based on RBFNs and to evaluate their efficiency compared to the traditionally used feedforward deep neural networks (DNNs). The research considers the primary types of operational IT risk events related to disruptions and failures of systems and equipment that ensure business operations. Method: A Gaussian function was used as the transformation function to build different RBFN models. The learning curves of the considered functions are provided to demonstrate the results of training and testing of the constructed RBFNs. Results and Discussion: The findings indicate that RBFNs can achieve an accuracy of up to 98%, making them comparable to multi-layer perceptrons but with fewer hidden layers and faster training times. The model performed well in determining the criticality of IT risk events, offering both transparency and practicality for implementation in the banking sector. Research Implication: The obtained findings show that the use of RBFNs in the indication system will provide a high-quality determination of the degree of impact of operational IT risks on the banking sector.