Australian PII Detection Model
Fine-tuned GLiNER for 18 Australian PII types.
Model Performance
- Training F1 : 100%
- Test Accuracy : 91.7%
- Parameters : 195,175,936
- Training Time : 17.9 minutes
Supported Entities
- PERSON_NAME
- TAX_FILE_NUMBER
- MEDICARE_NUMBER
- ABN
- ACN
- DRIVER_LICENSE
- PASSPORT_NUMBER
- PHONE_NUMBER
- EMAIL_ADDRESS
- PHYSICAL_ADDRESS
- DATE_OF_BIRTH
- BSB_ACCOUNT_NUMBER
- CREDIT_CARD_NUMBER
- SUPER_FUND_MEMBER_NUMBER
- SALARY_AMOUNT
- EMPLOYER_NAME
- HEALTH_FUND_NUMBER
- IP_ADDRESS
Document Types Supported
- Tax Returns
- Bank Statements
- Superannuation Statements
- Insurance Documents
- Employment Records
Training Details
- Base: urchade/gliner_medium-v2.1
- Data: 2000 synthetic AU documents
- Epochs: 10
- Batch: 8
- LR: 5e-6
- GPU: Google Colab T4
Limitations
- Trained on synthetic data only
- Real world accuracy may vary
- Needs OCR for scanned PDFs
- Use with regex for best results
Legal Notice
Research purposes only. Always include human review for production PII redaction.
- Downloads last month
- 8