Hybrid AI-Driven Proactive Detection of Dormant Cyber-Attacks in Sudanese Digital Banking Networks

Abstract

Detecting dormant cyber-attacks before they cause harm remains a major concern for Sudan’s digital banking sector, especially with the scarcity of accessible, real-world banking data. In this work, a hybrid artificial intelligence system was designed and tested to address this challenge. The solution blends traditional anomaly detection techniques such as Isolation Forest and Local Outlier Factor—with multi-layer perceptron (MLP) neural networks, combining their strengths through an integrated decision layer. To create a meaningful testbed, synthetic datasets were built to mimic authentic transaction patterns and dormant attack scenarios, using local market insights as a guide. Trials with this system yielded promising outcomes. The hybrid approach reached a detection accuracy of 93% and a recall rate of 91%, while reducing false alarms to just 4%. These improvements not only surpassed traditional models, which struggled to exceed 85% accuracy, but also lessened the burden of false alerts on security teams by over 60%. Such results suggest that hybrid AI methods can offer substantial benefits for banks operating in environments where real data is limited.Going forward, the study encourages Sudanese banks and research bodies to collaborate in building real banking datasets and pilot-testing intelligent security systems. Such joint efforts will be vital to advancing the security and reliability of digital banking in the country.