Scholarlink

<p>The rapid proliferation of Internet of Things (IoT) devices has introduced new challenges in edge computing, particularly regarding security. As IoT devices are often resource-constrained and operate in decentralized environments, traditional security mechanisms are inadequate to safeguard sensitive data and prevent malicious attacks. This paper introduces a Novel Adaptive Epsilon-Greedy Reinforcement Learning (NAEGRL) method to enhance the security of IoT-Edge computing. The proposed method employs reinforcement learning with an adaptive approach, utilizing an epsilon-greedy algorithm to dynamically balance exploration and exploitation, thereby enabling the system to continuously learn and adapt to evolving threats. The proposed security is adaptive, and hence, whenever it finds some malicious packets in the underlying IoT-Edge communication, the security layer immediately identifies and discards these malicious packets and also becomes adapted to such malicious packets by strengthening the security layer with an updated policy. Due to its adaptivity, the proposed security model enhances its knowledge database, allowing it to fully prevent similar or different malicious attacks in the future and maintain uninterrupted data communication even if the same external attack occurs at any time in the future. The NAEGRL method intelligently selects optimal defense strategies in real-time, minimising the attack surface and responding to novel attack vectors. The proposed security method uses the Epsilon (ℇ) value, which is calculated based on attack volume and packet metadata. The adaptivity of the proposed methodology results in a minimal requirement of exploration rather than exploitation of an unknown IoT-Edge environment, which brings fast communication in the network in a minimal time duration. The proposed security method is designed to prevent DDoS and/or Adversarial attacks. Such attacks are the most prevalent in modern IoT ecosystems. Besides Adaptivity, the proposed security method is more scalable and robust than other ML-based security methods. Extensive experiments conducted on Simulation, Public, and In-house IoT datasets demonstrate that NAEGRL achieves a detection accuracy of 99.37%, surpassing benchmark methods in terms of detection rate, resilience, scalability, and robustness. These results highlight NAEGRL as an effective and adaptive solution for securing next-generation IoT-Edge systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reinforcement learning based security method for IoT-Edge computing

  • 期刊论文

  • 发表日期：

    2026年3月30日

查看全文
  • Anit Kumar,
  • Dhanpratap Singh

摘要

The rapid proliferation of Internet of Things (IoT) devices has introduced new challenges in edge computing, particularly regarding security. As IoT devices are often resource-constrained and operate in decentralized environments, traditional security mechanisms are inadequate to safeguard sensitive data and prevent malicious attacks. This paper introduces a Novel Adaptive Epsilon-Greedy Reinforcement Learning (NAEGRL) method to enhance the security of IoT-Edge computing. The proposed method employs reinforcement learning with an adaptive approach, utilizing an epsilon-greedy algorithm to dynamically balance exploration and exploitation, thereby enabling the system to continuously learn and adapt to evolving threats. The proposed security is adaptive, and hence, whenever it finds some malicious packets in the underlying IoT-Edge communication, the security layer immediately identifies and discards these malicious packets and also becomes adapted to such malicious packets by strengthening the security layer with an updated policy. Due to its adaptivity, the proposed security model enhances its knowledge database, allowing it to fully prevent similar or different malicious attacks in the future and maintain uninterrupted data communication even if the same external attack occurs at any time in the future. The NAEGRL method intelligently selects optimal defense strategies in real-time, minimising the attack surface and responding to novel attack vectors. The proposed security method uses the Epsilon (ℇ) value, which is calculated based on attack volume and packet metadata. The adaptivity of the proposed methodology results in a minimal requirement of exploration rather than exploitation of an unknown IoT-Edge environment, which brings fast communication in the network in a minimal time duration. The proposed security method is designed to prevent DDoS and/or Adversarial attacks. Such attacks are the most prevalent in modern IoT ecosystems. Besides Adaptivity, the proposed security method is more scalable and robust than other ML-based security methods. Extensive experiments conducted on Simulation, Public, and In-house IoT datasets demonstrate that NAEGRL achieves a detection accuracy of 99.37%, surpassing benchmark methods in terms of detection rate, resilience, scalability, and robustness. These results highlight NAEGRL as an effective and adaptive solution for securing next-generation IoT-Edge systems.