Phishing Attack Detection on Websites Using Machine Learning

Phishing attacks usually copy reliable websites, such as banks and financial institutions, in an attempt to obtain personal information, such as passwords and credit card details. This study suggests a hybrid phishing detection system that combines the Back Propagation Neural Network (BPNN) for classification with XGBoost for feature selection. For training (80%) and testing (20%), a dataset of 11,000 URLs was employed, including both phishing and authentic samples. Important URL-based characteristics were extracted, including URL length, discrepancy character, HTTPS appearance, and domain age. High identification accuracy (97.5%), precision (96.8%), recall (98.2%), and F1-score (97.5%) were obtained by all systems. When compared to traditional classifiers (SVM, Random Forest), the proposed model shows better performance in identifying zero-day phishing efforts, Explanation measures were used to assess the model, and Scikit-LARN was used to simulate it in python. The results confirm that the algorithm can successfully detect and stop phishing efforts in real time.