Phishing Email Detection · Dissertation

Methods & tools

Findings

• Trained and evaluated models on a dataset of 18,634 emails
• Logistic Regression model: 95.4% accuracy, F1 score 0.942
• Fine-tuned DistilBERT on 1,000 samples: ~94-95% accuracy/recall
• Designed a FastAPI-based pipeline for serving predictions, documented as the dissertation's proposed deployment architecture