AI- Powered Behavioural Biometrics for Fraud Detection in Digital Banking: A Next-Generation Approach to Financial Cybersecurity
Isaac Adinoyi Salami, Anuoluwapo Deborah Popoola, Michael Olayinka Gbadebo, Faith Hauwa Oluwapamilerin Kolo, Temilade Oluwatoyin Adesokan-Imran
Abstract
This study investigates the limitations of traditional fraud detection techniques in digital banking and explores the applicability of AI-powered behavioral biometrics as a next-generation solution for enhancing cybersecurity. Using publicly available datasets, including the PaySim Financial Transactions Dataset, Credit Card Fraud Detection Dataset, and HMOG Dataset, this research applies machine learning models such as Random Forest, Long Short-Term Memory (LSTM) networks, and Convolutional Neural Networks (CNNs). These models were evaluated using quantitative metrics including Accuracy, Precision, Recall, F1 Score, and AUC-ROC. The LSTM network demonstrated superior performance, achieving 97.9% accuracy, 95.6% precision, and 93.4% recall, outperforming other models. The results reveal that deep learning frameworks significantly enhance fraud detection efficiency, minimize false positives, and improve prediction accuracy. Furthermore, the use of publicly available datasets enhances the study’s reproducibility and transparency. Ethical considerations related to privacy, user consent, and algorithmic accountability are also discussed, highlighting the importance of responsible AI deployment in digital banking systems. This research aims to address evolving cybersecurity threats by integrating advanced deep learning models with behavioral biometrics for real-time anomaly detection. The findings demonstrate the effectiveness of AI models in accurately detecting complex fraud patterns and propose practical recommendations for integrating such systems within existing digital banking infrastructures. Recommendations include improving algorithmic transparency, establishing ethical guidelines, and investing in infrastructure upgrades to facilitate seamless implementation. This work offers a valuable foundation for future research aimed at developing robust and adaptive fraud detection systems that prioritize both efficiency and ethical compliance.