Scholarlink

<p>A thorough understanding of intrusion detection systems (IDS) is essential for network security, as these systems detect known and unknown attacks through traffic analysis. Traditional approaches such as K-Nearest Neighbors (KNN) and Support Vector Machines (SVM) have performed well but are limited by their dependence on manual feature extraction, which restricts scalability on large datasets. In this article, we propose the BAT framework (Boosted Attention-enhanced Tabular learning), a structured multi-stage pipeline for network intrusion detection. BAT integrates an attention-based feature weighting mechanism that suppresses irrelevant input signals and amplifies discriminative features, followed by an XGBoost classifier that serves as the core classification engine. The framework is trained and evaluated on the NSL-KDD benchmark dataset, and compared against CNN, SVM, Decision Tree, and Logistic Regression baselines. The proposed BAT framework achieves superior classification accuracy of 96%, effectively combining feature-level attention with gradient-boosted ensemble learning to improve detection precision, reduce manual feature engineering, and handle diverse attack categories including DoS, Probe, R2L, and U2R.</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

Detection of Network Irregularities Using Machine Learning Techniques

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Merin Susan Philip,
  • Theegala Sindhu,
  • Devu Sai Sathwika,
  • Guduru Dinesh,
  • Krishna Chaitanya Rayasam

摘要

A thorough understanding of intrusion detection systems (IDS) is essential for network security, as these systems detect known and unknown attacks through traffic analysis. Traditional approaches such as K-Nearest Neighbors (KNN) and Support Vector Machines (SVM) have performed well but are limited by their dependence on manual feature extraction, which restricts scalability on large datasets. In this article, we propose the BAT framework (Boosted Attention-enhanced Tabular learning), a structured multi-stage pipeline for network intrusion detection. BAT integrates an attention-based feature weighting mechanism that suppresses irrelevant input signals and amplifies discriminative features, followed by an XGBoost classifier that serves as the core classification engine. The framework is trained and evaluated on the NSL-KDD benchmark dataset, and compared against CNN, SVM, Decision Tree, and Logistic Regression baselines. The proposed BAT framework achieves superior classification accuracy of 96%, effectively combining feature-level attention with gradient-boosted ensemble learning to improve detection precision, reduce manual feature engineering, and handle diverse attack categories including DoS, Probe, R2L, and U2R.