Benchmarking Traditional ML Approaches in Phishing URL Detection

Phishing attacks continue to pose a major cybersecurity challenge by exploiting deceptive URLs to obtain sensitive information. Although deep learning approaches such as CNNs, RNNs, and Transformers have demonstrated state of the art detection performance, traditional machine learning classifiers remain widely utilized because of their efficiency,interpretability, and relatively low resource requirements. In this study, we implement and evaluate ten machine learning models including Logistic Regression, Gradient Boosting, CatBoost, XGBoost, and Multi-Layer Perceptron on a publicly available phishing URL dataset from Kaggle. Experimental results show that ensemble based models,particularly XGBoost and Random Forest, achieve the highest accuracy, while Logistic Regression offers competitive performance with the advantages of simplicity, interpretability, and low computational overhead. The findings highlight the tradeoffs between accuracy, interpretability, and computational cost,providing practical guidance for selecting appropriate models in real world phishing detection systems.