Proactive Zero Day Threat Detection in 5G Mobile Edge Computing Using Transformer Driven Pre Execution Deep Learning Framework Process

Miracle Educational Society Group of Institutions, Andhra Pradesh, India

^* Corresponding author email: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 20 November 2025

Abstract

The deployment of 5G Mobile Edge Computing (MEC) at a tremendous speed has significantly increased the attack surface making zero-day vulnerabilities more palatable to exploit by the adversaries. Intrusion detection systems are signature-dependent, with anomaly identification being post-execution, leading to high latency and ineffectiveness at the adaptation against new threats. It is clearly stated that the need for practical, before-execution detection frameworks becomes paramount given the ultra-low latency environment and computational budget of MEC. This paper describes a Transformer-based pre-execution anomaly detection framework that is meant primarily for zero-day attack mitigation in 5G edge nodes. The framework is based on five novel methods that enhance accuracy, efficiency, and interpretability. The Pre-Execution Tokenized Executed Plan Transformer (PETEP-Trans) is an early-stage semantic intent-sequence model for zero-day prevention. Causal Residual Attribution Maps (CRAMs) provide interpretable causal localization of anomalies. The Drift-Calibrated Conformal Risk Control (DCRC) error evaluates the detection results after distributional drift. Latency-Optimal Token Skipping and Early Exit (LOTSEE) play with the gate to either keep the computation going because the machine already recognizes that the detection is successful or early-exit to save processing time. The last method, Shadow-Execution Hypersim Transformer (SEHT), accurately measures the performance of the newly crafted models and readily prepares for lack standard deviation by applying shadow execution, thereby increasing resilience against obfuscated attacks. All combined evaluations leave the integrated pipeline reaching 96% and above in zero-day detection accuracy and good scope to set false-positive rates below 2%, a range of 40% lower latency in many cases while judiciously providing interpretability with adaptability in dynamic MEC environment. Hence, these are therefore ascertaining the new reality in the realm of real-time, explainable, and efficient security at the edge by quantum leaps for facing the challenge of zero-day threats over their weakness in the next-generation mobile infrastructure in the making process.