Stacking-Based Model for Phishing Website Detection System

Phishing attacks pose a significant threat in cybersecurity and therefore their detection is of critical importance to ensure the security of digital environments. Machine learning methods have proven to be an effective approach to detect phishing websites. In this study, PyCaret library, a powerful tool among automatic machine learning libraries, is used to implement various machine learning algorithms quickly and efficiently. The primary objective is to develop high-performance machine learning models to detect phishing websites using PyCaret. For this purpose, PhiUSIIL dataset is used. Multiple machine learning algorithms were analyzed on the dataset using PyCaret library, the best performing algorithms were identified and further optimized through hyperparameter tuning. In addition, an ensemble model was created using the stacking method and the performance was improved. The analysis revealed that the Logistic Regression algorithm achieved 99.98% accuracy, the K-Nearest Neighbors algorithm 99.74% accuracy, and the Naive Bayes algorithm 99.24% accuracy. Later, a new model created with the stacking method using these 3 models increased the accuracy rate to 99.99%, demonstrating the effectiveness of the study in detecting phishing websites.