Abstract Cyberattacks have been rising steadily since the 1990s, and today the manufacturing and industrial sectors have become prime targets. U. S. manufacturing is seen as a lucrative target because of its rich space for exploitation, the fear of production halts, and the lack of a reliable, self-sufficient supply chain that can support operations during crises. With the growing use of interconnected technologies, entry points for attackers are more numerous than ever. Traditional methods such as signature-based or static defenses have proven ineffective, while artificial intelligence (AI) -driven approaches have shown promise but often lack consistency, performing well in some areas while failing in others. This study addresses that challenge by examining existing AI models used for cyber threat detection, evaluating their advantages and limitations, and proposing a more reliable alternative. This paper proposes a lightweight and easily deployable Deep Hybrid Learning (DHL) model trained and tested on the TONIoT dataset. The model was compared against ten of the most widely used machine learning (ML) and deep learning (DL) models in cybersecurity and achieved superior performance with 98. 13% accuracy, 98. 82% precision, 98. 24% recall, and 98. 53% F1. This study provides practical recommendations to strengthen industrial systems and protect manufacturing enterprises from the growing wave of cyber threats.
Maghanaki et al. (2026) studied this question.