An Adaptive Deep Reinforcement Learning Framework for Real-Time Identity Verification and Threat Response in Networked Systems

Abstract

Networked systems increasingly connect users, devices, applications, and services, making continuous identity verification essential for maintaining secure digital environments. Adaptive security technologies can learn from changing behavioral patterns and respond to emerging threats more effectively than static mechanisms. This study addresses the problem of maintaining accurate real-time identity verification while dynamically responding to evolving identity-related threats in networked systems. Static authentication methods and fixed security rules may fail when legitimate behavior changes or attackers develop new strategies, increasing false decisions and delaying threat mitigation. To address this challenge, this study develops an adaptive deep reinforcement learning framework that integrates identity, behavioral, device, and network information into dynamic security states. The framework enables a learning agent to select verification and threat-response actions according to current conditions and continuously improve its policy through environmental feedback. Experimental results show that the framework achieves 97.2% verification accuracy, 96.5% precision, 96.8% recall, and a 96.6% F1-score with an average response time of 38 ms. These results demonstrate improved performance compared with traditional machine learning, deep learning, reinforcement learning, and existing DRL approaches. The findings extend existing research by jointly optimizing identity verification and threat response through adaptive decision-making, providing a more responsive approach for evolving network security environments.