AI-Based Threat Monitoring Framework for Critical Infrastructure in Developing Economies

Abstract

The increasing digitalization of critical infrastructure by developing countries has opened new security problems that existing conventional methods have failed to address. Particularly with advanced threats rapidly evolving. This paper aims to develop a system that monitors both active and passive threats using Artificial Intelligence integrated with lightweight deep learning for optimization to watch for threats in places where resources are limited. We used an hybrid model that combines Autoencoder and LSTM. The model performed excellently in learning threats. We used 20 epochs in training the system and observed a stable convergence when training the epochs and when it was tested, which means it performed good in terms of it generalization capacity. The reconstruction error distributions result showed a significant separation between benign and anomalous events (p < 0.01). Our tests showed that the proposed threat model framework was very good at detecting threats with accuracy of 96% and a confidence range of 94.95% to 97.05%. We did a a two-sample t-test comparing reconstruction errors between normal and attack traffic that produced a statistically significant separation (t ≈ 18.7, p < 0.001) with threat detection score of 95%. This showed the model performed well at telling the difference between attack traffic. All these results together show that our model is reliable and can work well in places where resources are limited. The outcome is a system that can watch for threats and adapt to situations, within countries that are resource constrained.