Scholarlink

<p>Phishing websites continue to pose substantial threats to internet security, with conventional detection approaches including blacklisting and SSL certificate-based methods demonstrating limited effectiveness against evolving attack vectors. This paper presents a hybrid classification framework that integrates convolutional neural networks (CNN) for automated feature extraction with support vector machines (SVM) for binary classification, where SVM hyperparameters are optimized using an Adaptive Bat Algorithm (ABAT). A systematic evaluation of five bio-inspired optimization algorithms was conducted to determine optimal SVM parameter tuning approaches. Experimental validation was performed using datasets sourced from UCI Machine Learning Repository. The proposed method is achieved 99.90% accuracy, 100% precision, 99.83% recall, 99.91% F-score, and 0.0% false positive rate on the test dataset. Ablation studies confirm the individual contribution of each component, while confusion matrix analysis demonstrates perfect identification of malicious websites with zero false negatives. The hybrid approach significantly outperforms individual machine learning techniques and existing optimization algorithms, representing a 7.05% improvement over baseline CNN performance. These findings contribute to advancing automated cybersecurity defense mechanisms and provide a robust framework for real-time phishing detection applications.</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

Effective phishing website detection using CNN and SVM with adaptive hyperparameter optimization via BAT algorithms

  • 期刊论文

  • 发表日期：

    2025年12月12日

查看全文
  • Santosh Kumar Birthriya,
  • Priyanka Ahlawat,
  • Ankit Kumar Jain

摘要

Phishing websites continue to pose substantial threats to internet security, with conventional detection approaches including blacklisting and SSL certificate-based methods demonstrating limited effectiveness against evolving attack vectors. This paper presents a hybrid classification framework that integrates convolutional neural networks (CNN) for automated feature extraction with support vector machines (SVM) for binary classification, where SVM hyperparameters are optimized using an Adaptive Bat Algorithm (ABAT). A systematic evaluation of five bio-inspired optimization algorithms was conducted to determine optimal SVM parameter tuning approaches. Experimental validation was performed using datasets sourced from UCI Machine Learning Repository. The proposed method is achieved 99.90% accuracy, 100% precision, 99.83% recall, 99.91% F-score, and 0.0% false positive rate on the test dataset. Ablation studies confirm the individual contribution of each component, while confusion matrix analysis demonstrates perfect identification of malicious websites with zero false negatives. The hybrid approach significantly outperforms individual machine learning techniques and existing optimization algorithms, representing a 7.05% improvement over baseline CNN performance. These findings contribute to advancing automated cybersecurity defense mechanisms and provide a robust framework for real-time phishing detection applications.