Scholarlink

<p>With the growing sophistication of cyber-attacks, robust and adaptive network intrusion detection systems (NIDS) are essential for ensuring cybersecurity. Conventional NIDS models often struggle with high false positive rates and limited real-time detection capabilities, necessitating advanced deep learning solutions. This study presents an optimized network intrusion detection framework leveraging a hybrid BiLSTM–temporal convolutional network (BiLSTM–TCN) model, optimized with a sine–cosine algorithm with adaptive learning (SCA–AL) for hyperparameter tuning. The proposed framework consists of four key stages: (1) Feature extraction using principal component analysis and statistical feature engineering to enhance data representation, (2) feature selection via least absolute shrinkage and selection operator (LASSO) to eliminate redundant attributes, (3) classification using a bidirectional long short-term memory (BiLSTM) network coupled with temporal convolutional networks, which effectively captures both long-range dependencies and local temporal patterns in network traffic data, and (4) hyperparameter optimization using SCA–AL, which dynamically adjusts search parameters to improve convergence efficiency and reduce computational overhead. Experimental evaluation of CICIDS-2017 and TON-IoT datasets demonstrates that the BiLSTM–TCN model achieves an accuracy of 99.89%, outperforming traditional CNN–LSTM architectures. Performance assessment based on F1-score, precision, recall, and detection rate highlight the model’s effectiveness in cyber-attack mitigation. The results validate the framework’s potential to provide a scalable, real-time, and high-accuracy intrusion detection system for modern cybersecurity challenges.</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

High-performance intrusion detection using hybrid BiLSTM–TCN model with adaptive sine–cosine algorithm for hyperparameter optimization

  • 期刊论文

  • 发表日期：

    2025年10月30日

查看全文
  • Emad Alsuwat,
  • Hatim Alsuwat

摘要

With the growing sophistication of cyber-attacks, robust and adaptive network intrusion detection systems (NIDS) are essential for ensuring cybersecurity. Conventional NIDS models often struggle with high false positive rates and limited real-time detection capabilities, necessitating advanced deep learning solutions. This study presents an optimized network intrusion detection framework leveraging a hybrid BiLSTM–temporal convolutional network (BiLSTM–TCN) model, optimized with a sine–cosine algorithm with adaptive learning (SCA–AL) for hyperparameter tuning. The proposed framework consists of four key stages: (1) Feature extraction using principal component analysis and statistical feature engineering to enhance data representation, (2) feature selection via least absolute shrinkage and selection operator (LASSO) to eliminate redundant attributes, (3) classification using a bidirectional long short-term memory (BiLSTM) network coupled with temporal convolutional networks, which effectively captures both long-range dependencies and local temporal patterns in network traffic data, and (4) hyperparameter optimization using SCA–AL, which dynamically adjusts search parameters to improve convergence efficiency and reduce computational overhead. Experimental evaluation of CICIDS-2017 and TON-IoT datasets demonstrates that the BiLSTM–TCN model achieves an accuracy of 99.89%, outperforming traditional CNN–LSTM architectures. Performance assessment based on F1-score, precision, recall, and detection rate highlight the model’s effectiveness in cyber-attack mitigation. The results validate the framework’s potential to provide a scalable, real-time, and high-accuracy intrusion detection system for modern cybersecurity challenges.