Scholarlink

<p>The rapid evolution of communication technologies has led to an explosive growth of connected devices, offering greater convenience and efficiency in individuals’ daily lives. However, these interconnected networks have also amplified security risks, allowing malicious entities to manipulate network characteristics. Due to their inherent complexity, such attacks often remain undetected and demand more intelligent detection systems for effective mitigation. In recent years, Machine Learning (ML) and Deep Learning (DL) frameworks have shown significant potential in identifying these emerging threats. Nonetheless, these models require further enhancement in selecting the most relevant traffic features to minimise issues and improve detection accuracy. To resolve this issue, this research paper introduces a novel Chaotic Evoked Tunicate Swarm Optimization-based Feature Selector (CETS-FS) and a Dynamic Pyramidal Transformer Network (DPTN) for detecting and classifying network attacks. The proposed framework comprises three key stages: (i) data preprocessing, (ii) feature selection using the CETS-FS technique, and (iii) attack classification through the DPTN network. In the first stage, Z-score normalization is applied to ensure the data follows a normal distribution. Subsequently, the Synthetic Minority Oversampling Technique (SMOTE) is employed to overcome class imbalance issues. The most significant features are then selected using the CETS-FS technique, followed by the application of the DPTN for precise attack classification. The proposed framework was evaluated using multiple benchmark datasets, comprising UNSW-NB15, NSL-KDD, and CICIDS-2017. The performance of the framework was validated by leveraging standard classification metrics like precision, accuracy, F1-score, specificity, recall and compared against several existing feature selection–integrated learning frameworks. Empirical results (averaged across NSL-KDD, UNSW-NB15, and CICIDS-2017) demonstrate that the proposed framework attained an average accuracy of 0.995, precision of 0.990, recall of 0.989, specificity of 0.992, and F1-score of 0.997, with a detection time of 0.754&#xa0;s. These outcomes confirm that the proposed hybrid framework reveals strong powers in rapidly detecting malicious traffic patterns, thereby enhancing the security of dynamic networked environments.</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

Integrating hybrid deep learning architecture with enhanced feature selection techniques to mitigate the multiple attacks

  • 期刊论文

  • 发表日期：

    2025年12月29日

查看全文
  • M. Saritha,
  • Saidireddy Malgireddy

摘要

The rapid evolution of communication technologies has led to an explosive growth of connected devices, offering greater convenience and efficiency in individuals’ daily lives. However, these interconnected networks have also amplified security risks, allowing malicious entities to manipulate network characteristics. Due to their inherent complexity, such attacks often remain undetected and demand more intelligent detection systems for effective mitigation. In recent years, Machine Learning (ML) and Deep Learning (DL) frameworks have shown significant potential in identifying these emerging threats. Nonetheless, these models require further enhancement in selecting the most relevant traffic features to minimise issues and improve detection accuracy. To resolve this issue, this research paper introduces a novel Chaotic Evoked Tunicate Swarm Optimization-based Feature Selector (CETS-FS) and a Dynamic Pyramidal Transformer Network (DPTN) for detecting and classifying network attacks. The proposed framework comprises three key stages: (i) data preprocessing, (ii) feature selection using the CETS-FS technique, and (iii) attack classification through the DPTN network. In the first stage, Z-score normalization is applied to ensure the data follows a normal distribution. Subsequently, the Synthetic Minority Oversampling Technique (SMOTE) is employed to overcome class imbalance issues. The most significant features are then selected using the CETS-FS technique, followed by the application of the DPTN for precise attack classification. The proposed framework was evaluated using multiple benchmark datasets, comprising UNSW-NB15, NSL-KDD, and CICIDS-2017. The performance of the framework was validated by leveraging standard classification metrics like precision, accuracy, F1-score, specificity, recall and compared against several existing feature selection–integrated learning frameworks. Empirical results (averaged across NSL-KDD, UNSW-NB15, and CICIDS-2017) demonstrate that the proposed framework attained an average accuracy of 0.995, precision of 0.990, recall of 0.989, specificity of 0.992, and F1-score of 0.997, with a detection time of 0.754 s. These outcomes confirm that the proposed hybrid framework reveals strong powers in rapidly detecting malicious traffic patterns, thereby enhancing the security of dynamic networked environments.