Phishing Email Detector — Fine-tuned Phi-3-mini
Performance
| Metric | Score |
|---|---|
| Accuracy | 99.00% |
| Precision | 98.23% |
| Recall | 100.00% |
| F1 Score | 99.11% |
Training Details
- Base model: Phi-3-mini-4k-instruct (3.8B params)
- Method: LoRA fine-tuning (r=16, alpha=16)
- Trainable params: 29.8M (0.78%)
- Training samples: 9,000
- Epochs: 2, Final loss: 1.1593
- Hardware: Tesla T4 15.6GB
- Training time: 231 minutes
Datasets
- CEAS_08, SpamAssassin, Nazario, Enron
- Total: 74,388 balanced emails
Novel Contribution
Explainable phishing detection with risk score (0-100), triggered element explanation, and confidence level.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for omerfarooq223/phishing-detector-phi3-lora
Base model
microsoft/Phi-3-mini-4k-instruct