Efficient Phishing Detection System Using Boosting and Feature Engineering Techniques

Main Article Content

Sivaraju Kuraku
Naveen Shrivastava
Mahesh Marri
Khaja Moiezuddin Mohammed
Rashmika Jalagam
Sridevi Allu

Abstract

Phishing schemes have been a constant challenge to internet security, which requires strong detection strategies. This study is the best phishing detection system that utilises the UCI Repository's Phishing Websites Data Set, which includes 11,055 occurrences and 30 characteristics of URLs, domains, and webpage content. The phishing and legitimate data, which went through an extensive preprocessing stage, including the elimination of noise, label encoding, and feature engineering to retrieve relevant URL-based, content-based, and third-party-derived features. The XGBoost model was trained to function as a classifier using gradient boosting and regularization methods. All of the following metrics ACC, PRE, REC, F1-score(F1), ROC, and SHAP show that the model performed exceptionally well: 99.8% accuracy, 99.9% precision, 99.5% recall, and 98.7% F1-score. The ROC curve and SHAP analysis illustrated that there were very strong discriminative features of the model, and those were DNS_Record, Mouse_Over, Web Forwards, Web Traffic, and URL Length. Comparative analysis with SVM and DNN models further confirmed XGBoost’s superior performance in accurately and reliably detecting phishing websites, establishing it as a robust, scalable, and interpretable solution for cybersecurity applications

Article Details

Section

Articles

How to Cite

Kuraku, S., Shrivastava, N., Marri, M., Mohammed, K. M., Jalagam, R., & Allu, S. (2024). Efficient Phishing Detection System Using Boosting and Feature Engineering Techniques. International Journal of Humanities and Information Technology, 6(04), 579-593. https://doi.org/10.21590/

Similar Articles

You may also start an advanced similarity search for this article.