Scholarlink

<p>The transmission of users’ personal data via networks poses an increased risk of being exposed to phishing threats. Phishing URLs hiding malicious web addresses have become a widespread means of compromising user privacy. This paper is to improve the accuracy and efficiency of detecting and defending against phishing attacks in multimedia applications, aiming to safeguard users'personal data from exposure via phishing URLs. The proposed approach can be summarized in three phases. In the first phase, cosine and Jaccard similarities between training and test datasets are calculated to obtain a comprehensive understanding of the similarities of data points. A hybrid similarity score is built by combining these two measures in the second phase of the experiment to effectively blend content-related and quantity-related similarities. The third and final phase involves training a model on XGBoost that uses a hybrid similarity score as an input and then tunning up its hyperparameters using a genetic algorithm with the optimization of classification performance. Finally, the created powerful classifiers are merged to form a stacked ensemble model, increasing the overall accuracy of the classification. The strenuous test evaluation of 11,429 rows, 88 columns, through key metrics, such as Precision, Recall, F-Score, and Accuracy show that the proposed anti-phishing model has outperformance parameters in other models. With this lean approach, accuracy in classification can be improved by the incorporation of similarity-based techniques and machine learning and evolutionary optimization, in which defense can be performed against phishing threats in multimedia environments. The new innovation involved in the proposed work is based on stacking an ensemble of similarity-based techniques with a learning and optimization framework of machines, ensuring robustness toward phishing threat defense.</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

Innovative anti-phishing framework using machine learning approach with evolutionary optimization to secure multimedia applications

  • 期刊论文

  • 发表日期：

    2025年4月8日

查看全文
  • Bhawna Sharma,
  • Savita Khurana,
  • Gaurav Sharma,
  • Sunita Rani,
  • Sudesh Kumari Nandal,
  • Pragya Chandi,
  • Rashi Rastogi,
  • Ángel Kuc Castilla,
  • Nitin Goyal

摘要

The transmission of users’ personal data via networks poses an increased risk of being exposed to phishing threats. Phishing URLs hiding malicious web addresses have become a widespread means of compromising user privacy. This paper is to improve the accuracy and efficiency of detecting and defending against phishing attacks in multimedia applications, aiming to safeguard users'personal data from exposure via phishing URLs. The proposed approach can be summarized in three phases. In the first phase, cosine and Jaccard similarities between training and test datasets are calculated to obtain a comprehensive understanding of the similarities of data points. A hybrid similarity score is built by combining these two measures in the second phase of the experiment to effectively blend content-related and quantity-related similarities. The third and final phase involves training a model on XGBoost that uses a hybrid similarity score as an input and then tunning up its hyperparameters using a genetic algorithm with the optimization of classification performance. Finally, the created powerful classifiers are merged to form a stacked ensemble model, increasing the overall accuracy of the classification. The strenuous test evaluation of 11,429 rows, 88 columns, through key metrics, such as Precision, Recall, F-Score, and Accuracy show that the proposed anti-phishing model has outperformance parameters in other models. With this lean approach, accuracy in classification can be improved by the incorporation of similarity-based techniques and machine learning and evolutionary optimization, in which defense can be performed against phishing threats in multimedia environments. The new innovation involved in the proposed work is based on stacking an ensemble of similarity-based techniques with a learning and optimization framework of machines, ensuring robustness toward phishing threat defense.