Phishing websites remain difficult to block because many campaigns use newly registered domains that have not yet reached reputation lists. This paper reviews four supervised learning methods commonly used for phishing-website detection: Decision Tree, Random Forest, Support Vector Machine, and Logistic Regression. The review focuses on studies using URL and domain features and compares the reported accuracy, precision, recall, and F1-score alongside practical concerns such as inference cost and interpretability. Across the reviewed evidence, Random Forest usually gives the strongest overall detection performance, while Logistic Regression offers a smaller and faster model for constrained devices. However, the numerical results are not directly interchangeable because the studies use different dataset versions, feature definitions, validation procedures, and tuning choices. The paper therefore treats the reported values as comparative evidence rather than results from a new experiment. It also discusses concept drift, adversarial manipulation, reproducibility, and the need for temporal evaluation. The main finding is that model selection should reflect deployment constraints as well as benchmark accuracy.

Phishing Detection Machine Learning Cybersecurity Random Forest Support Vector Machine Decision Tree Logistic Regression URL Features