Scholarlink

<p>The rapid proliferation of internet services and the growing complexity of cyber intrusions have outpaced the capabilities of traditional intrusion detection algorithms. An effective intrusion detection strategy adopts a comprehensive approach to safeguard critical systems from unauthorized access and potential attacks. However, network devices generate vast amounts of high-dimensional data, complicating the accurate detection of both known and unknown attacks. This research presents a comprehensive security solution for network intrusion detection that utilizes a machine learning (ML) approach, specifically employing a stacking ensemble framework alongside ensemble feature selection technique. The proposed technique incorporates dynamic feature ranking, utilizing Gaussian Naïve Bayes as the meta-classifier. To enhance performance and reduce overfitting, feature scaling and thorough hyperparameter optimization are employed. The performance is assessed using following intrusion detection datasets: NSL-KDD, UNSW_NB-15, CIC-IDS-2017 and ToN-IoT. The proposed stacking ensemble model outperforms state-of-the-art techniques, achieving 98.4% accuracy on NSL-KDD and 100% on ToN-IoT, demonstrating strong potential for real-world intrusion detection.</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 autonomous intrusion detection framework leveraging multi-layered stacking ensemble architectures with dynamic machine learning-driven feature subset optimisation

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Faraj Chishti,
  • Geetanjali Rathee

摘要

The rapid proliferation of internet services and the growing complexity of cyber intrusions have outpaced the capabilities of traditional intrusion detection algorithms. An effective intrusion detection strategy adopts a comprehensive approach to safeguard critical systems from unauthorized access and potential attacks. However, network devices generate vast amounts of high-dimensional data, complicating the accurate detection of both known and unknown attacks. This research presents a comprehensive security solution for network intrusion detection that utilizes a machine learning (ML) approach, specifically employing a stacking ensemble framework alongside ensemble feature selection technique. The proposed technique incorporates dynamic feature ranking, utilizing Gaussian Naïve Bayes as the meta-classifier. To enhance performance and reduce overfitting, feature scaling and thorough hyperparameter optimization are employed. The performance is assessed using following intrusion detection datasets: NSL-KDD, UNSW_NB-15, CIC-IDS-2017 and ToN-IoT. The proposed stacking ensemble model outperforms state-of-the-art techniques, achieving 98.4% accuracy on NSL-KDD and 100% on ToN-IoT, demonstrating strong potential for real-world intrusion detection.