A binary multiobjective hippopotamus optimization algorithm for feature selection in phishing website detection

Phishing website detection remains a major challenge in cybersecurity as attackers continuously develop new techniques to deceive users. Identifying the most informative features from large datasets is essential to improve classification accuracy while reducing computational complexity. Feature selection is therefore widely addressed using metaheuristic optimization techniques due to their flexibility and global search capability. In this study, we propose a Binary Multiobjective Hippopotamus Optimization Algorithm (B-MOHOA) for feature selection in phishing website detection. The proposed method simultaneously optimizes two conflicting objectives: maximizing classification accuracy and minimizing the number of selected features. Unlike many existing studies that mainly focus on transfer functions, this work investigates the impact of binarization rules by combining a single transfer function with four different binarization strategies. Experiments are conducted on two phishing datasets, D1 and D2, using K-Nearest Neighbor (KNN) and Random Forest (RF) classifiers. On D1, the proposed approach achieves 96.02% accuracy with KNN and 97.38% with RF while reducing the feature space. On D2, it reaches 97.10% accuracy with only 8 features using KNN and 99.00% accuracy with 20 features using RF. The results demonstrate that B-MOHOA can effectively produce compact feature subsets while maintaining very high detection performance, highlighting the importance of binarization strategies in multiobjective feature selection for phishing detection.