Abstract—Phishing websites continue to pose significant threats to cybersecurity by deceiving users into revealing sensitive information. To address the complexity of highdimensional URL features and the challenge of identifying the most informative attributes, this paper proposes a machine-learning-based phishing detection model using a hybrid Feature Selection (FS) method that combines the Dragonfly Algorithm (DA) and the Whale Optimization Algorithm (WOA). This hybrid FS approach effectively removes irrelevant attributes, reduces model complexity, and improves the robustness of the learning process. The proposed model leverages the ISCX-URL2016 dataset, with adaptive boosting (AdaBoost [AB]) serving as the classifier and its hyperparameters optimized via grid search. Experimental results show that the union of DA and WOA (DA∪WOA) with AB outperforms existing methods, attaining 97.07% accuracy, 96.89% recall, 97.15% precision, 97.02% F1-score, and 94.14% Matthews Correlation Coefficient (MCC). The combination of hybrid FS and optimized classification not only boosts accuracy but also enhances computational efficiency, making the approach well-suited for real-time phishing detection systems.