Scholarlink

<p>Securing the information traversing network infrastructures from increasingly sophisticated attackers is paramount. Cybersecurity has evolved to become a critical and integral component of any organizational information system. To this end, the development of Intrusion Detection Systems (IDSs) for monitoring, analyzing, and identifying malicious activities within network data flows is crucial. This paper introduces five innovative IDS models that leverage the capabilities of Convolutional Neural Networks (CNNs). The foremost model, termed the Three-layer Convolutional Neural Network (TCNN), has been designed to minimize parameters and computational demands, thereby expediting the model training process. Complementing this, four additional models have been adopted employing the principles of Transfer Learning (TL): VGG16, VGG19, ResNet50, and ResNet152. We have subjected these models to rigorous evaluation using the NSL-KDD dataset, encompassing four categories of cyber-attacks—Probe, Remote to User (R2U), User to Root (U2R), and Denial of Service (DoS)—as well as benign network behavior. Our models are adept at conducting both binary and multi-class classifications across diverse training and testing data proportions. They have achieved impeccable scores, with precision, recall, and F1-score each reaching 100%. Concurrently, the detection rate, classification accuracy, and false alarm rate stand at 99.99%, 99.81%, and 0.004, respectively. These outcomes corroborate the efficacy of the proposed systems in assisting cybersecurity professionals to efficiently pinpoint network breaches. Furthermore, we present a comparative analysis between our models and selected cutting-edge IDS frameworks to underscore their superior performance and dominance in the field.</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

Optimized Convolutional Neural Network Frameworks for Automatic Intrusion Detection Systems in Wireless Cybersecurity Applications

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Youssef F. Sallam,
  • Samy Abd El-Nabi,
  • Walid El-Shafai,
  • Hossam El-din H. Ahmed,
  • Adel Saleeb,
  • Nirmeen A. El-Bahnasawy,
  • Fathi E. Abd El-Samie

摘要

Securing the information traversing network infrastructures from increasingly sophisticated attackers is paramount. Cybersecurity has evolved to become a critical and integral component of any organizational information system. To this end, the development of Intrusion Detection Systems (IDSs) for monitoring, analyzing, and identifying malicious activities within network data flows is crucial. This paper introduces five innovative IDS models that leverage the capabilities of Convolutional Neural Networks (CNNs). The foremost model, termed the Three-layer Convolutional Neural Network (TCNN), has been designed to minimize parameters and computational demands, thereby expediting the model training process. Complementing this, four additional models have been adopted employing the principles of Transfer Learning (TL): VGG16, VGG19, ResNet50, and ResNet152. We have subjected these models to rigorous evaluation using the NSL-KDD dataset, encompassing four categories of cyber-attacks—Probe, Remote to User (R2U), User to Root (U2R), and Denial of Service (DoS)—as well as benign network behavior. Our models are adept at conducting both binary and multi-class classifications across diverse training and testing data proportions. They have achieved impeccable scores, with precision, recall, and F1-score each reaching 100%. Concurrently, the detection rate, classification accuracy, and false alarm rate stand at 99.99%, 99.81%, and 0.004, respectively. These outcomes corroborate the efficacy of the proposed systems in assisting cybersecurity professionals to efficiently pinpoint network breaches. Furthermore, we present a comparative analysis between our models and selected cutting-edge IDS frameworks to underscore their superior performance and dominance in the field.