ENHANCED PHISHING DETECTION THROUGH HYBRID MACHINE LEARNING AND URL ANALYSIS

Abstract

Phishing attacks represent a significant and highly dangerous form of cybercrime that occurs on the
internet. The project leverages a dataset containing phishing URLs. These URLs are the web addresses used by
cybercriminals to carry out phishing attacks. To detect and combat these phishing attempts effectively, the project
employs a variety of machine learning algorithms. These include decision tree, linear regression, random forest,
naive Bayes, gradient boosting classifier, K-neighbors classifier, support vector classifier, and a hybrid model
referred to as LSD. In addition to using these algorithms, the project incorporates advanced techniques such as
cross-fold validation and Grid Search Hyperparameter Optimization. To determine how well these models work, the
project uses specific evaluation metrics. These metrics include precision, accuracy, recall, and F1-score. In
enhancing the Phishing Detection System, a Stacking Classifier with RF + MLP using LightGBM exhibits improved
performance.