An ensemble learning approach for detecting phishing URLs in encrypted TLS traffic

Phishing is a fraudulent method used by hackers to acquire confidential data from victims, including security passwords, bank account details, debit card data, and other sensitive data. Owing to the increase in internet users, the corresponding network attacks have also grown over the last decade. Existing phishing detection methods are implemented for the application layer and are not effectively adapted to the transport layer. In this paper, we propose a novel phishing detection method that extends beyond traditional approaches by utilizing a multi-model ensemble of deep neural networks, long short term memory, and Random Forest classifiers. Our approach is distinguished by its unique feature extraction from transport layer security (TLS) 1.2 and 1.3 network traffic and the application of advanced deep learning algorithms to enhance phishing detection capabilities. To assess the effectiveness of our model, we curated datasets that include both phishing and legitimate websites, using features derived from TLS 1.2 and 1.3 traffic. The experimental results show that our proposed model achieved a classification accuracy of 99.61%, a precision of 99.80%, and a Matthews Correlation Coefficient of 99.22% on an in-house dataset. Our model excels at detecting phishing Uniform Resource Locator at the transport layer without data decryption. It is designed to block phishing attacks at the network gateway or firewall level.