Comparative evaluation of machine learning algorithms for phishing site detection

Area of Science:

• Cybersecurity
• Machine Learning
• Deep Learning

Background:

• The rise of internet technologies has led to increased electronic transactions and a surge in phishing attacks targeting sensitive user information.
• Phishing poses a significant threat, aiming to steal confidential data for financial gain or identity theft.
• Existing anti-phishing methods require enhancement for effective phishing URL identification.

Purpose of the Study:

• To evaluate the performance of eight machine learning (ML) and deep learning (DL) algorithms in identifying phishing URLs.
• To compare the efficacy of algorithms including SVM, KNN, RF, DT, XGBoost, LR, and CNN.
• To identify the most effective DL/ML model for robust anti-phishing technology.

Main Methods:

• Utilized two real-world datasets (Mendeley and UCI) for model training and evaluation.
• Employed performance metrics such as accuracy, precision, recall, False Positive Rate (FPR), and F1-score.
• Implemented feature engineering, SMOTE for class imbalance, and rigorous hyperparameter tuning for all models, including a novel CNN approach.

Main Results:

• The Convolutional Neural Network (CNN) model achieved superior accuracy in phishing URL detection compared to other evaluated algorithms.
• All tested ML and DL models showed consistent performance across both datasets, indicating stability.
• The study highlights the effectiveness of CNN in combating phishing threats.

Conclusions:

• Machine learning and deep learning, particularly CNN, offer promising solutions for advanced anti-phishing technologies.
• The developed models provide reliable and stable performance in identifying phishing URLs.
• Further research and implementation of CNN-based systems can significantly enhance user security against online fraud.