REAL-TIME PHISHING WEBSITE DETECTION USING GRADIENT BOOSTING AND ENSEMBLE STRATEGIES

Abstract

Phishing attacks remain one of the most prevalent and damaging cybersecurity threats, targeting users through deceptive websites designed to steal sensitive information such as login credentials and financial data. Traditional blacklist-based detection systems often fail to identify newly generated phishing URLs, making real-time and intelligent detection mechanisms essential. This paper proposes a Real-Time Phishing Website Detection Framework leveraging Gradient Boosting and ensemble learning strategies to enhance classification accuracy and detection speed. The system extracts lexical, host-based, and contentbased features from URLs and web pages, which are then processed using advanced Gradient Boosting algorithms such as XGBoost and LightGBM. To improve robustness and generalization, multiple ensemble strategies, including stacking and voting classifiers, are employed. The proposed model is optimized for real-time deployment with low latency and high throughput. Experimental evaluation demonstrates superior detection accuracy, reduced false positive rates, and improved responsiveness compared to traditional machine learning models. The framework provides a scalable and effective solution for proactive phishing prevention in modern web environments.