Tools
Tools
Omotola, Akinola David (2024) Using Genetic Algorithms for Optimized Feature Selection in Machine Learning and Deep Learning Models to Detect Phishing Websites. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (3MB) | Preview |
Abstract
Given the advancement of network technology and the global rise in internet usage, cybersecurity concern has also risen. Phishing attacks which use misleading tactics to leak sensitive information bring about significant financial losses, reputational harm, system disruptions and legal consequences for both victim and organisation. This growing threat demonstrates the critical need for more robust detection systems, especially when accounting for evolving attack tactics. This study addresses the issue by investigating the use of a Genetic Algorithm (GA) combined with Support Vector Machine (SVM), Random Forest Classifier (RFC), Gradient Boost Classifier (GBC) and Graph Neural Network (GNN) to identify phishing websites. The findings of the experiment show that hybrid models perform better than single-base models, providing higher accuracy, F1-score, and AUC on a small dataset. According to this research, the GA-GNN and GA-RFC were the best and most improved model with accuracy of 95%, F1-score of 94.74% and AUC of 95.45%. The paper does admit several limitations such as, performance drop with larger sample size. Nonetheless, the study highlights the promise of hybrid models in tackling phishing detection problems and raising detection precision.
Actions (login required)
 |
View Item |