A Hybrid Approach for Alluring Ads Phishing Attack Detection Using Machine Learning.

Area of Science:

• Cybersecurity
• Machine Learning
• Data Science

Background:

• Phishing attacks are increasingly sophisticated, posing significant threats to internet users.
• Existing machine learning approaches for phishing detection often rely on limited datasets and URL features.

Purpose of the Study:

• To develop and evaluate a modern, machine learning-based approach for web phishing detection.
• To propose an efficient layered classification model utilizing URL, text, and image features.

Main Methods:

• A large dataset of 20,000 website URLs was compiled, extracting 22 features per URL.
• Natural Language Processing (NLP) techniques were applied to website text, including Optical Character Recognition (OCR) for text within images.
• A layered classification model was implemented using Support Vector Machine (SVM), XGBoost, Random Forest, Multilayer Perceptron, Linear Regression, Decision Tree, Naïve Bayes, and Support Vector Classification (SVC) algorithms.

Main Results:

• The XGBoost algorithm demonstrated superior performance, achieving 94% accuracy in training and 91% in testing.
• Multilayer Perceptron, Random Forest, and Decision Tree models also showed high accuracy (91%, 91%, and 90% respectively in testing).
• Text-based classification using Logistic Regression and SVM achieved accuracies of 87% and 88% respectively.

Conclusions:

• The proposed layered classification model effectively detects sophisticated phishing websites using URL, text, and image features.
• Machine learning algorithms, particularly XGBoost, offer a powerful solution for accurate and efficient phishing detection.
• This research enhances internet user security by enabling early detection of advanced phishing attacks.