Hybrid Deep Learning Framework for Real-Time Phishing Website Detection and Cybersecurity Enhancement

Abstract

Phishing sites are still one of the most dangerous online threats as they lure people to hand over their confidential details like passwords, bank information or other sensitive personal info. Conventional methods of catching phishing scanners mainly rely on a set of rules, and features designed manually and because of this often miss very recent and sophisticated phishing attacks. Authors of this paper first designed a Hybrid Deep Learning Structure for Real-Time Phishing Website Detection and Cybersecurity Enhancement based on a combination of CNN and LSTM. The newly-developed mechanism tries to automatically recognize and locate spatial as well as sequential patterns in URLs, webpage contents, and security-related attributes for detecting Phishing websites from safe ones. Having a detection system working in real-time presents the possibility of quickly spotting harmful websites, which will ultimately minimize the risks of cyberattacks and at the same time protect the users effectively. Proposed model was tested against an established dataset for phishing websites, and its performance measures like accuracy precision recall, and F1-score were compared to other models. Experimental results show that performance measures of the new model are: accuracy 98.76%, precision 98.42%, recall 98.65%, and F1-score 98.53%, which are even better than those of standard machine learning models and deep learning models separately. Besides, the model even managed to keep the detection delay very low during the live examples. In short, this combination of deep learning approaches brings an effective, flexible, and automatically upgrading way to identify phishing, and a great step towards a safer digital world.