2024 · Academic research
Phishing URL detection
Exploring transformer-based models for classifying URLs, from dataset preparation to model evaluation.
- Context
- Academic machine-learning research
- My role
- Model development and evaluation
- Tools
- Python · PyTorch · Transformers

Exploring a classification problem
This project investigated encoder-decoder transformer models for phishing URL detection. The work covered data preparation, training scenarios, and evaluation of model behavior.
The work
I worked with Python and PyTorch to prepare URL data and experiment with transformer-based classification. The project included dataset balancing and hyperparameter-tuning scenarios.
Research as a starting point
Model evaluation is one part of a useful AI system. Operational use would also require an appropriate evaluation dataset, monitoring, an understanding of false positives and false negatives, and testing against changing inputs.
Evidence and boundaries
The public repository contains the research implementation. This is an academic project, not a deployed security service. Historical performance figures are intentionally omitted from this overview until their experimental conditions can be presented alongside them.