Detecting and Mitigating Cybersecurity Threats in Nigerian Commercial Banks Using Machine Learning: A Case Study of Selected Institutions

• Folakemi B. OYEYEMI

  Author

• Usman A. ABDURRAHMAN

  Author

Nigerian commercial banks are facing rising cybersecurity threats, with electronic fraud losses surpassing ₦5.1 billion in 2023 [1], [2]. Traditional signature-based detection systems are reactive and struggle against dynamic, advanced attacks. This study developed and tested a machine learning-based cybersecurity threat detection and automated mitigation system designed for Nigerian commercial banks, with GTBank and Access Bank as institutional case studies [3], [4]. Guided by the CRISP-DM methodology [5], [6], [7], the study built and tested three classifiers: Random Forest, Support Vector Machine (SVM), and Long Short-Term Memory (LSTM), across two benchmark datasets, the CICIDS 2017 network intrusion dataset and the IEEE-CIS Financial Fraud Detection dataset. To correct pronounced class imbalance without data leakage into the test set, SMOTE oversampling was restricted to the training partition only. The results showed that all models exceeded the minimum macro F1-score of 0.90 for network intrusion detection. Random Forest achieved near-perfect classification (macro F1: 0.9974), followed by LSTM (0.9468) and SVM (0.9203). For financial fraud detection, which presented extreme class imbalance (3.5% fraud rate), Random Forest achieved the highest macro F1-score (0.7616), while LSTM demonstrated the strongest fraud recall (0.76), highlighting its suitability for sequential threat analysis. These results are consistent with recent studies on ML-based fraud detection using imbalanced datasets [8], [9], [10], [11]. An automated mitigation module was integrated via a Flask RESTful API, enabling real-time responses. While direct experimental comparison with a deployed signature-based IDS was outside the scope of this study, the results provide strong evidence that the evaluated ensemble and deep learning models substantially exceed performance thresholds associated with rule-based detection systems documented in the literature [12], [13]. The findings therefore suggest that ensemble and deep learning models offer a superior alternative to traditional signature-based systems, providing a high-accuracy threat detection and response framework for the Nigerian financial sector.

• 39_Oyeyemi and Abdurrahman (2026)

Copyright (c) 2026 Folakemi B. OYEYEMI, Usman A. ABDURRAHMAN (Author)

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.