Scholarlink

With internet usage continuing to surge, phishing attacks are becoming a very concerning phenomenon when it comes to safeguarding sensitive organizational and individual users’ information. Such attacks use fake malicious URLs in an attempt to steal the user’s sensitive information. They are said to cause losses worth $17,700 per minute as stated by CSO Online. This paper seeks to overcome such a challenge by examining the performance of voting, stacking, and bagging ensemble learning approaches for the purpose of phishing URL detection on our data with 35 features extracted. We assess these techniques based on their classification accuracy and F1-Score using URL-based, lexical-based, and domain-based features. Empirical results indicate that ensemble methods considerably outperform single classifiers wherein the stacking classifier achieved the best results, with an accuracy of 93.71% and F1-Score of 93.79%.

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

Ensemble Approaches for Phishing URL Detection: A Deep Dive into Advanced Machine Learning Methods

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Aditi Rathi,
  • Himanshi Kaushal,
  • Manushree Grover,
  • Ritika Kumari

摘要

With internet usage continuing to surge, phishing attacks are becoming a very concerning phenomenon when it comes to safeguarding sensitive organizational and individual users’ information. Such attacks use fake malicious URLs in an attempt to steal the user’s sensitive information. They are said to cause losses worth $17,700 per minute as stated by CSO Online. This paper seeks to overcome such a challenge by examining the performance of voting, stacking, and bagging ensemble learning approaches for the purpose of phishing URL detection on our data with 35 features extracted. We assess these techniques based on their classification accuracy and F1-Score using URL-based, lexical-based, and domain-based features. Empirical results indicate that ensemble methods considerably outperform single classifiers wherein the stacking classifier achieved the best results, with an accuracy of 93.71% and F1-Score of 93.79%.