Phishing Email Detection Using NLP

AbstractPhishing emails use deceptive messages, impersonation, and misleading links to persuade recipients to disclose credentials,open malicious attachments, or perform unauthorised actions. Traditional filtering methods often rely on known senderreputation, rules, and signatures; these remain useful but may not recognise newly worded or targeted messages. This paperproposes a Natural Language Processing (NLP)-based phishing email detection system that analyses message text andselected email metadata to classify messages as likely phishing or legitimate. The proposed pipeline includes safe messageingestion, text normalisation, tokenisation, feature extraction using TF-IDF, and supervised classification with models suchas Logistic Regression and Linear Support Vector Machines. Optional URL and header features can complement textualsignals. A review dashboard presents a prediction with confidence and evidence for human verification rather than treating amodel output as a definitive verdict. The paper defines an evaluation plan using precision, recall, F1-score, false-positive rate,and confusion-matrix analysis. No experimental results are claimed because the proposed system has not been tested in acontrolled experiment. Privacy, dataset leakage prevention, adversarial wording, and responsible handling of email contentare considered throughout the design.