An Original Contribution to Continual Learning Theory for Adaptive Cybersecurity Systems in Nigerian Threat Landscapes

Cybersecurity threats in Nigerian digital environments evolve rapidly, with new attack vectors, malware variants, and social engineering techniques emerging faster than static ML-based detection models can be retrained. Continual learning, which enables AI models to acquire new knowledge without catastrophic forgetting of prior capabilities, offers a theoretical solution, but existing continual learning approaches have not been adapted to cybersecurity detection tasks in African threat contexts. This study develops an original contribution to continual learning theory for adaptive cybersecurity systems in Nigerian threat landscapes. Three theoretical contributions are presented: a Nigerian Cybersecurity Task Stream formalisation that models the evolution of cyber threats as a non-stationary sequence of detection tasks with measurable covariate shift, based on 24 months of threat intelligence data from five Nigerian financial institutions and the CERRT-NG national CSIRT; an original elastic weight consolidation variant (EWC-Threat) that selectively preserves detection capability for high-severity threat categories during model updates while allowing plasticity for new threat types; and a replay memory construction policy that prioritises the retention of adversarial edge cases from prior threat streams most likely to recur in future cycles, based on a recurrence probability model trained on Nigerian threat intelligence data. The combined framework was evaluated on a 24-month cybersecurity event dataset, demonstrating 11 percent higher detection retention across threat class transitions compared to vanilla EWC, with comparable new task acquisition. The study constitutes an original contribution to continual learning theory for security applications.

Keywords: continual learning, cybersecurity, adaptive AI, Nigeria, threat detection