Abstract & Overview
This studyevaluates the Cybersecurity Threat Intelligence Framework for Nigerian Financial Institutions (CTIF-NG). CTIF-NG is a five-layer architecture that brings together supervised and unsupervised machine learning algorithms, big-data analytics pipelines, and sector-specific threat-indicator feeds calibrated to documented Nigerian financial threat taxonomies. It is not a generic framework retrofitted for Nigeria it was designed around the CBN (2022), NDPA (2023), ISO/IEC 27035-2, and NIST SP 800-150 requirementsthat Nigerian institutions actually face. To evaluate the classification engine at its core, we constructed a structured simulation dataset of 20,000 records across seven threat classes. Feature distributions were drawn from published NSL-KDD (Tavallaee et al., 2009) and CICIDS-2017 (Sharafaldin et al., 2018) benchmark statistics and adapted to the Nigerian financial threat taxonomy. Real scikit-learn v1.8.0 experiments on the proposed RF and GBM and MLP Soft-Vote Ensemble produced: detection accuracy of 99.30%, macro-precision of 98.42%, macro-recall of 98.18%, F1-score of 98.30%, Matthews Correlation Coefficient of 0.9896, and AUC-ROC of 0.9999. All seven baseline classifiers were outperformed. APT class recall came in at 95.0% the lowest of all classes which was anticipated given that APT features were deliberately configured to overlap with normal traffic. Scenario projections suggest CTIF-NG can reduce mean time-to-detect by 64.8% and mean time-to-respond by 71.3% compared to conventional SIEM-only baselines, though live SOC validation is still required before these figures can be relied upon operationally