PREDICTION PHISHING ATTACK OVER INTERNET OF THINGS ENVIRONMENT USING MACHINE LEARNING AND OPTIMIZATION APPROACH

Abstract

Phishing websites remain a major cyber-security threat because they imitate trusted online services to steal sensitive user information. This study proposes a learning framework for phishing website detection using publicly available data and phishing related features. The dataset is divided into training and testing subsets to learn discriminative patterns between phishing and legitimate websites. A Support Vector Regression (SVR) classifier is employed, and its hyper-parameters are optimized using the BaldEagle Optimizer (BEO) to improve generalization and reduce false positives. Experimental results indicate that the proposed BEO-optimized SVM achieved the highest classification accuracy of 95%. In comparison, the baseline methods obtained lower accuracies, including 90% for conventional SVM, 88% for Neural Network, 83% for Logistic Regression, 83% for Decision Tree, and 82% for AP. These findings suggest that BEO-based hyper-parameter optimization improves phishing website classification performance on the selected benchmark dataset.