An Ensemble Machine Learning Framework for Robust Phishing Website Detection

Phishing attacks pose a serious threat to cybersecurity since they are always changing to take advantage of gullible people and compromise private data. This study investigates the use of deep learning (DL) and machine learning (ML) techniques for phishing website detection. Using the Phishing Websites dataset from OpenML, which has 11,055 instances with 30 different features, we apply a thorough strategy that includes model construction, performance evaluation, and data preprocessing. We thoroughly evaluate four different algorithms: Multi-Layer Perceptron (MLP), Random Forest, XGBoost, and Logistic Regression. Stratified dataset partitioning, categorical variable encoding, and feature scaling are all included in the preparation procedure. Confusion matrices and ROC curve visualizations are added to the accuracy, precision, recall, F1-score, and ROC-AUC metrics used in performance measurement. According to experimental results, Logistic Regression provides a solid basis with 92.8% accuracy. While Random Forest and MLP show improved performance, XGBoost performs exceptionally well with 97.7% accuracy and 97.9% F1-score. These results highlight the usefulness of ensemble boosting techniques, especially XGBoost, in building reliable phishing detection systems for real-world cybersecurity applications.