Illegal web platforms known as phishing websites adopted high-risk threats to mimic legitimate online platforms in order to steal important data of users, like login credentials and financial information. These forged platforms often involve internet tactics to attract unsuspecting victims through URL manipulation. Many recent machine learning techniques are proposed for detecting web attackers. However, under an increasing number of internet users, they struggle to reveal recent phishing strategies for stealing sensitive information covered by fake websites. This study presents a comprehensive methodology for detecting phishing websites through a proposed hybrid technique (RF_PSO) by involving the Random Forest (RF) classification algorithm with Particle Swarm Optimisation (PSO) as a hyperparameter optimisation for tuning parameters of the RF and how complex and Explainable Artificial Intelligence (XAI) using Local Interpretable Model-Agnostic Explanations (LIME). Experiments were conducted on three phishing datasets: Zieni, UCI Phishing Websites, and Mendeley to assess the proposed model’s robustness and generalization. The proposed RF_PSO achieved superior performance, with accuracies of 90.3% (Zieni), 97.1% (UCI), and 98.7% (Mendeley), and corresponding ROC-AUC scores of 0.9653, 0.9955, and 0.9989, respectively. Eight metrics evaluated the models, including accuracy, precision, recall, F1 score, MCC, AUC, and reported classification, along with the training and testing times, which were used to comprehensively evaluate the models. This work offers open research for exploring hybrid swarm intelligence methods to detect the more recently evolving phishing strategies.
