Edge-Enabled Cyber Threat Detection Using Compressed Transformer Architectures

Abstract

The exponential growth in digital connectivity has significantly expanded the attack surface for cyber threats, demanding more intelligent and real-time detection mechanisms. Traditional cybersecurity systems often rely on centralized processing, which introduces latency and limits responsiveness, particularly in time-critical or bandwidth-constrained environments. This research proposes an innovative edge-enabled cyber threat detection framework utilizing compressed transformer architectures to overcome these challenges. By deploying lightweight yet powerful models directly at the edge of the network, this approach enables rapid, local analysis of complex and heterogeneous threat data. To support this work, we utilize the Cyber Threat Dataset: Network, Text & Relation from Kaggle, which integrates structured network traffic, unstructured textual alerts, and relational entity data to simulate attack scenarios. Our methodology leverages transformer-based models for both natural language processing and structured data representation, with model compression techniques such as pruning, quantization, and knowledge distillation applied to optimize performance for edge deployment. The compressed models maintain high detection accuracy while drastically reducing computational overhead, making them ideal for deployment in resource-constrained environments. Experimental results demonstrate the effectiveness of the proposed system in identifying sophisticated threats with minimal latency and high throughput. This research highlights the potential of combining edge computing with advanced deep learning to achieve scalable, efficient, and real-time cyber threat detection—laying the groundwork for the next generation of intelligent, decentralized cybersecurity systems.