Scholarlink

<p>A growing body of current research highlights intrusion detection as a pivotal area of study within Internet of Things (IoT) networks, demonstrating its potential to enhance operational efficiency. The high diversity of IoT devices, their protocols and standards, and their limited computational resources have led to the appearance of novel security challenges. However, traditional security solutions to intrusion detection systems (IDS) are often unable to counter evolving new attack patterns on an overwhelming volume of records with insufficient feature diversity. The advent of increased computational power, coupled with accelerated network data traffic and reduced computing costs, has enabled researchers across disciplines to use deep learning (DL) techniques in IDS datasets to provide high accuracy and real-time detection. In this context, first, this study applies Kernel Principal Component Analysis (KPCA) to reduce the dimensionality of the dataset while preserving meaningful class distinctions. Subsequently, the Lévy flight mechanism is employed to identify functional feature sets to increase diversity in the search process. In addition, this work integrates deep learning models and addresses prevailing gaps in existing intrusion detection research, focusing on their classification performance and robustness against vulnerabilities. Additionally, we optimize Deep Neural Networks with Long Short-Term Memory architecture (DNN-LSTM) using the adaptive Lévy flight Grasshopper Optimization Algorithm (GOA) to fine-tune hyperparameters, improving anomaly detection efficacy. The proposed approach is evaluated using the CIC-IDS 2017, TON-IoT, and NSL-KDD datasets, with experimental results that indicate that the DNN-LSTM model significantly improves detection accuracy, particularly for novel or small-sample attack scenarios, outperforming standalone DL models. Empirical analysis further reveals that our approach surpasses state-of-the-art deep neural networks across several key performance metrics, underscoring its effectiveness in advancing intrusion detection research.</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

An effective hybrid deep learning metaheuristic model for robust IoT intrusion detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月19日

查看全文
  • Alok Kumar Shukla,
  • Shubhra Dwivedi,
  • Aishwarya Mishra

摘要

A growing body of current research highlights intrusion detection as a pivotal area of study within Internet of Things (IoT) networks, demonstrating its potential to enhance operational efficiency. The high diversity of IoT devices, their protocols and standards, and their limited computational resources have led to the appearance of novel security challenges. However, traditional security solutions to intrusion detection systems (IDS) are often unable to counter evolving new attack patterns on an overwhelming volume of records with insufficient feature diversity. The advent of increased computational power, coupled with accelerated network data traffic and reduced computing costs, has enabled researchers across disciplines to use deep learning (DL) techniques in IDS datasets to provide high accuracy and real-time detection. In this context, first, this study applies Kernel Principal Component Analysis (KPCA) to reduce the dimensionality of the dataset while preserving meaningful class distinctions. Subsequently, the Lévy flight mechanism is employed to identify functional feature sets to increase diversity in the search process. In addition, this work integrates deep learning models and addresses prevailing gaps in existing intrusion detection research, focusing on their classification performance and robustness against vulnerabilities. Additionally, we optimize Deep Neural Networks with Long Short-Term Memory architecture (DNN-LSTM) using the adaptive Lévy flight Grasshopper Optimization Algorithm (GOA) to fine-tune hyperparameters, improving anomaly detection efficacy. The proposed approach is evaluated using the CIC-IDS 2017, TON-IoT, and NSL-KDD datasets, with experimental results that indicate that the DNN-LSTM model significantly improves detection accuracy, particularly for novel or small-sample attack scenarios, outperforming standalone DL models. Empirical analysis further reveals that our approach surpasses state-of-the-art deep neural networks across several key performance metrics, underscoring its effectiveness in advancing intrusion detection research.