Phishing Detection System Through Hybrid Machine Learning Based on URL

Abstract

Phishing attacks on the internet by leveraging a comprehensive phishing URL-based dataset. Employing an array of machine learning algorithms, including Decision Tree [4], Linear Regression[4], Random Forest [4], Naive Bayes, Gradient Boosting Classifier, Support Vector Classifier, and a novel hybrid LSD model, the study aims to enhance cyber threat detection. Through meticulous cross-fold validation and Grid Search Hyper parameter Optimization, As an extension we have applied a hybrid model by combining the predictions of multiple individual models like Stacking Classifier, an ensemble technique, to combine predictions from Random Forest [4] Classifier[4] and MLP Classifier as base classifiers. It uses LGBM Classifier as a meta-estimator to make the final prediction, extending the project's capabilities for improved classification performance. Evaluation metrics such as precision, accuracy, recall, and F1-score are employed to assess model effectiveness. The results underscore the efficacy of the hybrid LSD model in mitigating phishing threats, providing a robust defense mechanism against evolving cyber threats. This research contributes to the advancement of cybersecurity measures and demonstrates the potential of machine learning in bolstering online security.