This artefact supports a final-year dissertation investigating phishing detection with a focus on explainable AI.
Many classifiers label emails, but few provide evidence that end users can understand.
Emails may contain sensitive data. PHISHLENS should avoid storing user inputs and should run locally for demos.
Adversarial wording and unseen scams can reduce performance. Future work: more datasets + stronger models.