Machine learning as a moderator in electronic banking fraud mitigation: evidence from South Africa’s banking sector

Abstract

The rapid integration of advanced technologies has reshaped banking operations, enhancing efficiency and customer service but simultaneously heightening vulnerability to fraud. In South Africa, electronic banking fraud has risen sharply, compelling financial institutions to adopt stronger risk management strategies. This study investigates how machine learning can reinforce the application of two key international frameworks, the COSO 2013 Internal Control Integrated Framework and the ISO 31000:2018 Risk Management Standard, in curbing E‑banking fraud. Employing a quantitative, cross‑sectional survey, data were gathered from 345 respondents through both paper‑based and online questionnaires, refined via pilot testing to ensure validity. Analytical procedures utilised Partial Least Squares Structural Equation Modelling (PLS‑SEM) in SmartPLS 4. Findings reveal that machine learning substantially enhances fraud mitigation and strengthens the effectiveness of both COSO and ISO standards when deployed together. COSO 2013 demonstrates a measurable impact on fraud reduction independently, yet its influence is magnified when integrated with machine learning. In contrast, ISO 31000:2018 shows limited effect unless complemented by machine learning techniques. Overall, the study underscores the critical value of aligning machine learning with established global risk management frameworks, thereby optimising fraud prevention, detection, and response within the banking sector.