Scholarlink

<p>The increasing prevalence of cybersecurity threats has amplified the essential for cuuting-edge network intrusion detection systems (NIDS) that ensure real-time protection and maintain data confidentiality, integrity, and accessibility. Existing IDS models struggle with weak interpretability, redundant features reduce efficiency, while imbalanced multi-class scenarios and limited generalization hinder accurate detection and large-scale deployment, especially for minority attacks. The proposed MMESF framework solves these problems by combining statistical and model-driven feature selection methods in an ensemble way. The proposed MMESF framework begins with SMOTE-Tomek sampling to take care of imbalance minority class and improve class representation. It then integrates multiple feature selection methods—including Pearson, ANOVA F-test, Chi-square, Mutual Information Gain, and RFE—combined through majority soft voting to determine the most relevant features. Finally, The Random Forest techniques attains high multiclass accuracy with a low false positive rate of 0.7-−0.8%, ensuring efficient generalization across three datasets with minimal computational cost. The proposed MMESF was experimented on the UNSW-NB15, CICIDS-2017, and CSE-CIC-IDS2018 datasets, achieving accuracies of 98.60%, 98.70%, and 98.40%, respectively. MMESF efficiently mitigates redundant features, enhances detection of minority attacks, and offers strong generalization for large-scale deployment with lower computational overhead.</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

Hybrid multi-model ensemble framework for cybersecurity: threat detection and defense against assorted network intrusion attacks

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Devraj Vishnu,
  • Maheshwari Prasad Singh

摘要

The increasing prevalence of cybersecurity threats has amplified the essential for cuuting-edge network intrusion detection systems (NIDS) that ensure real-time protection and maintain data confidentiality, integrity, and accessibility. Existing IDS models struggle with weak interpretability, redundant features reduce efficiency, while imbalanced multi-class scenarios and limited generalization hinder accurate detection and large-scale deployment, especially for minority attacks. The proposed MMESF framework solves these problems by combining statistical and model-driven feature selection methods in an ensemble way. The proposed MMESF framework begins with SMOTE-Tomek sampling to take care of imbalance minority class and improve class representation. It then integrates multiple feature selection methods—including Pearson, ANOVA F-test, Chi-square, Mutual Information Gain, and RFE—combined through majority soft voting to determine the most relevant features. Finally, The Random Forest techniques attains high multiclass accuracy with a low false positive rate of 0.7-−0.8%, ensuring efficient generalization across three datasets with minimal computational cost. The proposed MMESF was experimented on the UNSW-NB15, CICIDS-2017, and CSE-CIC-IDS2018 datasets, achieving accuracies of 98.60%, 98.70%, and 98.40%, respectively. MMESF efficiently mitigates redundant features, enhances detection of minority attacks, and offers strong generalization for large-scale deployment with lower computational overhead.