Scholarlink

<p>Cybersecurity faces persistent threats such as phishing, malware delivery, spam, and website defacement, making early detection of malicious URLs essential. Traditional approaches rely on manually engineered features that often miss subtle, evolving patterns in modern URL structures. We develop an Automated Feature Extraction Engine (AFEE) coupled with Feature Selection Engineering (FSE) to improve multiclass detection. AFEE derives lexical, statistical, structural, and semantic attributes from raw URLs; FSE then refines them using Chi-Square, Recursive Feature Elimination (RFE), Principal Component Analysis (PCA), Least Absolute Shrinkage and Selection Operator (LASSO), correlation filtering, and Local Interpretable Model-Agnostic Explanations (LIME). On a balanced dataset of 5,000 URLs across five classes (benign, malware, phishing, defacement, spam), we evaluate nine classifiers–Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), Logistic Regression (LR), Naïve Bayes (NB), AdaBoost (AB), and two ensembles (Hard and Soft Voting). In our experiments, GB, DT, and Soft Voting each achieved 99.91% accuracy. The results suggest that AFEE produces high-signal, low-redundancy features that favor tree-based decision boundaries, while probability averaging in Soft Voting further stabilizes predictions. Overall, the AFEE–FSE pipeline offers a scalable and reproducible approach to multiclass malicious-URL detection.</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

Automated AI framework for Malicious URL detection through enhanced feature extraction

  • 期刊论文

  • 发表日期：

    2025年11月20日

查看全文
  • Twinkle Patel,
  • Digvijaysinh Rathod

摘要

Cybersecurity faces persistent threats such as phishing, malware delivery, spam, and website defacement, making early detection of malicious URLs essential. Traditional approaches rely on manually engineered features that often miss subtle, evolving patterns in modern URL structures. We develop an Automated Feature Extraction Engine (AFEE) coupled with Feature Selection Engineering (FSE) to improve multiclass detection. AFEE derives lexical, statistical, structural, and semantic attributes from raw URLs; FSE then refines them using Chi-Square, Recursive Feature Elimination (RFE), Principal Component Analysis (PCA), Least Absolute Shrinkage and Selection Operator (LASSO), correlation filtering, and Local Interpretable Model-Agnostic Explanations (LIME). On a balanced dataset of 5,000 URLs across five classes (benign, malware, phishing, defacement, spam), we evaluate nine classifiers–Decision Tree (DT), Random Forest (RF), Gradient Boosting (GB), Support Vector Machine (SVM), Logistic Regression (LR), Naïve Bayes (NB), AdaBoost (AB), and two ensembles (Hard and Soft Voting). In our experiments, GB, DT, and Soft Voting each achieved 99.91% accuracy. The results suggest that AFEE produces high-signal, low-redundancy features that favor tree-based decision boundaries, while probability averaging in Soft Voting further stabilizes predictions. Overall, the AFEE–FSE pipeline offers a scalable and reproducible approach to multiclass malicious-URL detection.