A Novel Approach for Detecting Phishing Websites Using Hybrid Feature Selection and Ensemble Machine Learning

International Journal of Computer Science and Technology | Vol. 1, No. 5, May 2020 | pp. 33–40

Research Article

Title: A Novel Approach for Detecting Phishing Websites Using Hybrid Feature Selection and Ensemble Machine Learning

Names of Authors: Zainab Al-Husseini¹, Ahmed Al-Tamimi², and Noor Al-Kindi³

Authors’ Affiliations:
¹Department of Computer Science, University of Baghdad, Baghdad, Iraq
²Department of Software Engineering, King Fahd University of Petroleum and Minerals, Dhahran, Saudi Arabia
³Department of Information Systems, Sultan Qaboos University, Muscat, Oman

Abstract: Phishing attacks remain one of the most pervasive cyber threats, exploiting human vulnerabilities and deceptive visual elements to harvest sensitive user credentials and financial information. Attackers constantly refine URL obfuscation techniques and rapidly deploy ephemeral hosting domains, rendering traditional blacklisting databases ineffective against zero-day phishing campaigns. Machine learning classification models offer proactive defense mechanisms by analyzing lexical, structural, and content-based properties of suspect URLs. This study proposes a novel, high-accuracy phishing website detection system employing a hybrid feature selection pipeline coupled with an optimized ensemble classifier. The feature extraction phase analyzes 35 distinct lexical parameters from URLs, HTML source codes, and external domain registration metadata. To eliminate redundant or noisy variables, a hybrid feature selection mechanism combining Information Gain (IG) and Recursive Feature Elimination (RFE) is implemented, reducing the feature space down to the 18 most discriminative indicators. An ensemble stacking classifier comprising Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM) meta-learners is trained and validated on a balanced dataset of 40,000 legitimate and malicious URLs. Experimental evaluations indicate that the hybrid ensemble model achieves an overall classification accuracy of 98.6%, a false positive rate of 0.012, and a Mathews correlation coefficient (MCC) of 0.971, outperforming baseline individual classifiers. The proposed methodology provides a reliable, high-performance defense tool deployable as a real-time browser extension for internet users.

Keywords: Phishing detection, Machine learning, Feature selection, Ensemble classification, Cyber security, URL analysis

Manuscript Timeline: Received: February 15, 2020; Revised: March 22, 2020; Accepted: April 20, 2020; Published: May 1, 2020

Citation: Al-Husseini, Z., Al-Tamimi, A., & Al-Kindi, N. (2020). A novel approach for detecting phishing websites using hybrid feature selection and ensemble machine learning. International Journal of Computer Science and Technology, 1(5), 33–40. DOI: 10.46882/2020/IJCST/000005

Subscribe to read the full article: https://internationalscholarsjournals.org/subscribe-to-read