Scholarlink

<p>Intrusion Detection Systems (IDS) are essential for protecting networks from cyber threats and ensuring the security of critical infrastructures. This paper introduces an innovative hybrid model that combines&#xa0;Deep Belief Networks (DBNs),&#xa0;Principal Component Analysis (PCA), and&#xa0;Support Vector Machines (SVM)&#xa0;with&#xa0;adversarial learning&#xa0;to enhance the detection of both known and emerging cyber threats. The proposed model achieves a strong balance of accuracy, efficiency, and scalability, making it highly effective for real-world applications. By incorporating adversarial learning, the system can detect&#xa0;zero-day exploits—previously unseen attacks—and adapt to evolving threats. This is achieved by training the model on adversarial samples, which improves its resilience to variations in attack patterns. SVM further enhances the model’s ability to classify known and unknown threats with high precision. The model was rigorously tested on two benchmark datasets,&#xa0;NSL-KDD&#xa0;and&#xa0;CICIDS2017, achieving outstanding results. It demonstrated a high accuracy rate of&#xa0;99.73%&#xa0;on NSL-KDD and a low false positive rate of&#xa0;0.55%, while on CICIDS2017, it achieved a false positive rate of&#xa0;0.73%. The model also outperformed existing approaches in precision, recall, and F1-score, particularly in detecting complex attack types such as&#xa0;R2L (Remote-to-Local),&#xa0;U2R (User-to-Root), and&#xa0;Botnet attacks. Additionally, PCA reduced the feature space by&#xa0;40%, significantly lowering the model’s inference time to just&#xa0;9.0&#xa0;ms per sample. This makes the model highly suitable for real-time intrusion detection in high-traffic network environments, where speed and efficiency are critical. Overall, this hybrid model strengthens cybersecurity defenses and contributes to building resilient infrastructures. By effectively detecting and mitigating cyber threats, it supports the creation of safer, fairer, and more inclusive digital communities. The integration of adversarial learning and dimensionality reduction ensures robust, real-time protection against both known and emerging threats.</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

A Semi-Supervised Deep Learning Approach for Intrusion Detection and Classification for the Internet of Things

  • 期刊论文

  • 发表日期：

    2025年4月10日

查看全文
  • G. Uthradevi,
  • P. Thiruvasagam,
  • S. Mythili,
  • S. Oswalt Manoj

摘要

Intrusion Detection Systems (IDS) are essential for protecting networks from cyber threats and ensuring the security of critical infrastructures. This paper introduces an innovative hybrid model that combines Deep Belief Networks (DBNs), Principal Component Analysis (PCA), and Support Vector Machines (SVM) with adversarial learning to enhance the detection of both known and emerging cyber threats. The proposed model achieves a strong balance of accuracy, efficiency, and scalability, making it highly effective for real-world applications. By incorporating adversarial learning, the system can detect zero-day exploits—previously unseen attacks—and adapt to evolving threats. This is achieved by training the model on adversarial samples, which improves its resilience to variations in attack patterns. SVM further enhances the model’s ability to classify known and unknown threats with high precision. The model was rigorously tested on two benchmark datasets, NSL-KDD and CICIDS2017, achieving outstanding results. It demonstrated a high accuracy rate of 99.73% on NSL-KDD and a low false positive rate of 0.55%, while on CICIDS2017, it achieved a false positive rate of 0.73%. The model also outperformed existing approaches in precision, recall, and F1-score, particularly in detecting complex attack types such as R2L (Remote-to-Local), U2R (User-to-Root), and Botnet attacks. Additionally, PCA reduced the feature space by 40%, significantly lowering the model’s inference time to just 9.0 ms per sample. This makes the model highly suitable for real-time intrusion detection in high-traffic network environments, where speed and efficiency are critical. Overall, this hybrid model strengthens cybersecurity defenses and contributes to building resilient infrastructures. By effectively detecting and mitigating cyber threats, it supports the creation of safer, fairer, and more inclusive digital communities. The integration of adversarial learning and dimensionality reduction ensures robust, real-time protection against both known and emerging threats.