Original title: Robust phishing URL detection using FastText embeddings and deep sequential models
Shah Noor, Sibghat Ullah Bazai, Muhammad Imran Ghafoor, Alamgir Naushad +3 · September 2026
Phishing attacks trick people into giving away private data, money, and personal information. Old safety tools use lists of known harmful web addresses, or URLs. These lists cost a lot to maintain and fail to catch brand-new malicious links. To fix this, researchers tested seven deep learning models to spot phishing URLs automatically. They turned web addresses into numerical representations using a tool called FastText, which captures the fine parts of words. Then, they passed these representations into deep learning systems to learn patterns and classify the links. Across their tests, a model called Bidirectional Long Short-Term Memory, or BiLSTM, performed best. It read URL sequences in both forward and backward directions, reaching an accuracy of 98.62%. The system worked well across benchmark data and a separate external dataset, proving it can reliably detect hidden and unfamiliar phishing attacks.
Imagine you are a mail sorter. You have a list of known dangerous addresses. But clever scammers send letters with addresses that look almost real, like changing one letter. Your list cannot catch these slightly different addresses. Here, the system learns to spot suspicious patterns in the address itself, not just exact matches, by reading each address from both ends and understanding the characters.
Phishing attacks trick users into giving away passwords and bank details. Old detection methods cannot catch new or cleverly disguised URLs. This model automatically spots them with high accuracy, making online browsing safer for everyone.
The abstract notes that some prior studies achieved over 99% accuracy, though those used less robust setups. The current method achieves 98.62% and is validated on an external dataset, but its performance on never-before-seen URLs in real-world conditions is not explicitly tested.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Robust phishing URL detection using FastText embeddings and deep sequential models — Shah Noor, Sibghat Ullah Bazai, Muhammad Imran Ghafoor et al. — PeerJ Computer Science — doi:10.7717/peerj-cs.4074
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.