Hybrid Machine Learning Approach for Phishing URL Detection Using Stacking and LGBM Classifier

Abstract

Phishing attacks have become one of the most dangerous cybersecurity threats in modern internet environments. Cybercriminals create fraudulent websites and malicious URLs that imitate legitimate platforms to steal sensitive user information such as passwords, banking credentials, and personal details. Traditional phishing detection techniques such as blacklistbased systems and rule-based filtering methods are unable to detect newly generated phishing websites effectively. To overcome these limitations, machine learning approaches have gained significant importance in the field of cybersecurity. This research paper proposes a Hybrid Machine Learning Approach for Phishing URL Detection Using Stacking and LGBM Classifier to improve phishing website identification accuracy and reliability. The proposed system combines multiple machine learning algorithms through a stacking ensemble learning technique and utilizes the powerful LightGBM classifier as the meta-learning model. The implementation process includes URL dataset collection, preprocessing, feature extraction, feature optimization, model training, testing, and prediction generation. Important URL-based features such as URL length, subdomain count, HTTPS usage, suspicious symbols, and domain-related information are extracted and analyzed to identify phishing behavior. Multiple classifiers including Random Forest, Logistic Regression, Decision Tree, and Support Vector Machine are integrated to improve classification performance.