Scholarlink

<p>In today’s digital world, Internet of Things (IoT) networks face sophisticated threats, making intrusion detection and prevention crucial. Network intrusion detection systems (IDS) evaluate network traffic to identify harmful actions but struggle with accuracy and false positive rates (FPR). Effective IDSs must balance accuracy and FPR, but machine learning systems suffer from model drift and poor generalisation across datasets, hindering their adaptability to real-world cyberattacks.This study aims to construct a reliable and versatile IDS that recognises complex threats in real-time while maintaining high accuracy and low false positives. We propose an Optuna Variational Autoencoder With Double Deep Q-Network ( OPVAE-DDQN) adaptive IDS architecture using deep reinforcement learning agents. This integration enhances the comprehension of intricate data representations while streamlining decision-making processes for superior performance. Employing Optuna for hyperparameter optimization automates the tuning process, improving model flexibility across many activities and contexts. The VAE framework in OPVAE-DDQN facilitates the creation of varied data representations, enhancing state space exploration in reinforcement learning. This capacity is especially beneficial in constrained or uneven data settings, resulting in more resilient and generalizable models.Evaluations on the UNSW-NB15 and AWID datasets show our model outperforms state-of-the-art IDS in accuracy and FPR. On UNSW-NB15, OPVAE-DDQN achieved 91.25% accuracy, and on AWID, it achieved 98.36%. These results demonstrate the efficiency of our adaptive IDS in identifying emerging network threats.The proposed OPVAE-DDQN adaptive IDS significantly advances network security by providing a robust solution to evolving cyber threats. This research paves the way for developing more resilient intrusion detection 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

Smoke and mirrors: A variational optuna autoencoder with reinforcement learning for intrusion detection in IoT networks

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Isaac Osei Asante,
  • Libing Wu

摘要

In today’s digital world, Internet of Things (IoT) networks face sophisticated threats, making intrusion detection and prevention crucial. Network intrusion detection systems (IDS) evaluate network traffic to identify harmful actions but struggle with accuracy and false positive rates (FPR). Effective IDSs must balance accuracy and FPR, but machine learning systems suffer from model drift and poor generalisation across datasets, hindering their adaptability to real-world cyberattacks.This study aims to construct a reliable and versatile IDS that recognises complex threats in real-time while maintaining high accuracy and low false positives. We propose an Optuna Variational Autoencoder With Double Deep Q-Network ( OPVAE-DDQN) adaptive IDS architecture using deep reinforcement learning agents. This integration enhances the comprehension of intricate data representations while streamlining decision-making processes for superior performance. Employing Optuna for hyperparameter optimization automates the tuning process, improving model flexibility across many activities and contexts. The VAE framework in OPVAE-DDQN facilitates the creation of varied data representations, enhancing state space exploration in reinforcement learning. This capacity is especially beneficial in constrained or uneven data settings, resulting in more resilient and generalizable models.Evaluations on the UNSW-NB15 and AWID datasets show our model outperforms state-of-the-art IDS in accuracy and FPR. On UNSW-NB15, OPVAE-DDQN achieved 91.25% accuracy, and on AWID, it achieved 98.36%. These results demonstrate the efficiency of our adaptive IDS in identifying emerging network threats.The proposed OPVAE-DDQN adaptive IDS significantly advances network security by providing a robust solution to evolving cyber threats. This research paves the way for developing more resilient intrusion detection systems.