<![CDATA[This study addresses the growing threat of phishing attacks, particularly those exploiting malicious URLs that mimic legitimate websites and bypass traditional rule-based or blacklist detection systems. The purpose of this research is to conduct a comparative analysis and optimization of ensemble learning models for phishing URL detection using a combination of lexical and host-based features. The proposed approach involves feature extraction from URL structures and domain-related attributes, followed by classification using Random Forest and XGBoost. Hyperparameter optimization is performed using Grid Search to obtain the best model configuration. Model performance is evaluated using Stratified 5-Fold Cross-Validation, with metrics including accuracy and Area Under the ROC Curve (AUC), along with statistical significance testing. The experimental results show that the optimized Random Forest model achieves the best performance with an accuracy of approximately 0.97, slightly outperforming XGBoost with an accuracy of 0.96. Statistical analysis using a paired t-test indicates that this difference is significant (p = 0.000615). The results also demonstrate that combining lexical and host-based features significantly improves classification performance.This study provides empirical evidence that optimized ensemble learning models with hybrid feature representation can effectively detect phishing URLs with high accuracy and stability. The findings contribute to the development of more reliable and robust phishing detection systems in real-world cybersecurity environments.]]>
