Scholarlink

<p>While network security continues to evolve, efficient and scalable network intrusion detection systems (IDS) are more critical than ever. However, their deployment presents significant challenges due to their high computational and memory requirements. This resource burden often leads to performance bottlenecks and scalability issues in real-world scenarios. The present paper relies on a preprocessing method that transforms raw network traffic into grayscale images (28 × 28 pixels) and introduces a novel approach that leverages image texture features to achieve significant dimensionality reduction of network traffic data. Our approach achieves an accuracy of 99,97% and 99,85% for binary and multiclass classification respectively while reducing feature space from 784 to just 3 features, preserving essential characteristics and significantly lowering computational overhead. By integrating XGBoost as a classifier model, we show that high intrusion detection accuracy can be achieved with minimal computational resources. The approach has been tested on the InSDN dataset, suggesting its potential robustness for the detection of various attack classes. Experimental results reveal that our approach not only maintains competitive detection metrics but also appears to reduce processing overhead and memory requirements. Our findings suggest that image texture-based features efficiently bridge the gap between detection accuracy and resource efficiency in modern IDS implementations, providing a practical solution for resource-constrained environments.</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

Resource-Efficient Preprocessing for IDS: Dimensionality Reduction Through Image Texture Features

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Madjed Bencheikh Lehocine,
  • Hacene Belhadef,
  • Salim Djaaboub

摘要

While network security continues to evolve, efficient and scalable network intrusion detection systems (IDS) are more critical than ever. However, their deployment presents significant challenges due to their high computational and memory requirements. This resource burden often leads to performance bottlenecks and scalability issues in real-world scenarios. The present paper relies on a preprocessing method that transforms raw network traffic into grayscale images (28 × 28 pixels) and introduces a novel approach that leverages image texture features to achieve significant dimensionality reduction of network traffic data. Our approach achieves an accuracy of 99,97% and 99,85% for binary and multiclass classification respectively while reducing feature space from 784 to just 3 features, preserving essential characteristics and significantly lowering computational overhead. By integrating XGBoost as a classifier model, we show that high intrusion detection accuracy can be achieved with minimal computational resources. The approach has been tested on the InSDN dataset, suggesting its potential robustness for the detection of various attack classes. Experimental results reveal that our approach not only maintains competitive detection metrics but also appears to reduce processing overhead and memory requirements. Our findings suggest that image texture-based features efficiently bridge the gap between detection accuracy and resource efficiency in modern IDS implementations, providing a practical solution for resource-constrained environments.