A powerful machine learning method for detecting phishing threats - Article

Aug 4, 2026

Phishing threats leverage social engineering and deceptive web infrastructure to steal sensitive personal information by imitating legitimate websites. With rapid expansion of online services and rising cybercrime, detecting phishing websites remains a critical challenge. This study proposes a comprehensive machine learning approach using 48 discriminative features extracted from 10,000 webpages (5,000 phishing and 5,000 legitimate). Nine classifiers are evaluated, then ensemble models via soft voting and stacking are built, with soft voting achieving the best performance at 98.82% accuracy and F1 score.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Abstract

Introduction

Background and literature review

Proposed approach implementation

Results and evaluation

Conclusion

A powerful machine learning method for detecting phishing threats Mahmoud Baklizi1, Jamal Zraqou1, Mohammad Alkhazaleh2, Issa Atoum3, Faisal Alzyoud2, Musab B. Alzghoul2 1Department of Computer Science, Faculty of Information Technology, University of Petra, Amman, Jordan 2Department of Computer Science, Faculty of Information Technology, Isra University, Amman, Jordan 3Department of Software Engineering, Faculty of Information Technology, Philadelphia University, Amman, Jordan Article history: Received Nov 18, 2024 Revised Sep 13, 2025 Accepted Sep 27, 2025 Keywords: Cybersecurity Intrusion detection Machine learning Phishing detection Threat detection Corresponding Author: Phishing threats exploit social engineering and deceptive web infrastructure to steal sensitive personal information, often by mimicking legitimate websites. With the proliferation of online services and the increasing prevalence of cybercrime, detecting phishing websites has become a critical challenge. This study presents a comprehensive machine learning (ML)-based approach for detecting phishing websites. A total of 48 discriminative features were extracted from 10,000 webpages—comprising 5,000 phishing and 5,000 legitimate sites. Nine ML classifiers were initially evaluated, including random forest (RF), support vector machine (SVM), and XGBoost. Ensemble models based on soft voting and stacking were then constructed to improve detection performance. Among the models, the soft voting classifier (VC) achieved the best performance with an accuracy and F1-score of 98.82% . The results indicate that ensemble learning offers a robust solution for the automated detection of phishing websites. This is an open access article under the CC BY-SA license. Jamal Zraqou Department of Computer Science, Faculty of Information Technology, University of Petra Queen Alia Airport Street, Amman, Jordan Email: [Jamal.Zraqou@uop.edu.jo](Jamal.Zraqou@uop.edu.jo) Article Info ABSTRACT 1. INTRODUCTION The swift evolution of the internet, coupled with its expanding usage, complicates and furthers the security issue [1], [2] . Recently, numerous threats have emerged that aim to obtain sensitive personal information for financial gain or identity theft. One of these common threats is phishing. Phishing is a criminal activity that uses technology and social engineering to steal sensitive personal information from victims [3] . This information can include financial account details, login credentials, usernames, passwords, personal contacts, and social relationships [4] . In phishing, attackers contact users by phone, text, or email to solicit personal information while posing as well-known or respected businesses [4],[5] . Phishing websites often pretend to have urgent issues, such as unpaid invoices, suspicious account activity, or requests to log in to "verify" your password or account details. These phony websites can also request confidential information, including bank account numbers or credit card details. If you enter this information, cybercriminals can gain access to your accounts, steal your data, commit identity theft, and even infect your device with malware [6]-[8] . According to Awasthi and Goel [9], about half of cybersecurity experts noticed these. Due to the COVID-19 pandemic, they are subjected to attacks. In addition, there are around 1185 phishing attacks directed at enterprises every month, too. In turn, security professionals might have to use 1–4 days to fend off a cyber attack. Furthermore, about 30% of people in cybersecurity reported that phishing has a high rate of success now. Given the increase in phishing attacks, various methods to handle and mitigate them have been proposed. To address this persistent challenge, the present study initiates the development and testing of an effective machine learning (ML)-based system for identifying phishing sites through systematic page features. Whereas earlier works were

FAQs

What problem does the study address?

The study addresses phishing website detection, focusing on identifying deceptive sites that imitate legitimate webpages to steal sensitive information.

How is the dataset constructed for model training?

It uses 10,000 webpages, with 5,000 labeled as phishing and 5,000 as legitimate, from which 48 discriminative manual features are extracted.

Which modeling strategy produces the best results?

Ensemble learning is most effective, and the soft voting classifier achieves the highest performance with 98.82% accuracy and F1 score.

Aug 4, 2026

Upload by Clementine