ab-ai/pii_model

Personal Identifiable Information (PII Model)

This model is a fine-tuned version of bert-base-cased on the generator dataset. It achieves the following results:

• Training Loss: 0.003900
• Validation Loss: 0.051071
• Precision: 95.53%
• Recall: 96.60%
• F1: 96%
• Accuracy:99.11%

Model description

Meet our digital safeguard, a savvy token classification model with a knack for spotting personally identifiable information (PII) entities. Trained on the illustrious Bert architecture and fine-tuned on a custom dataset, this model is like a superhero for privacy, swiftly detecting names, addresses, dates of birth, and more. With each token it encounters, it acts as a vigilant guardian, ensuring that sensitive information remains shielded from prying eyes, making the digital realm a safer and more secure place to explore.

Model can Detect Following Entity Group

• ACCOUNTNUMBER
• FIRSTNAME
• ACCOUNTNAME
• PHONENUMBER
• CREDITCARDCVV
• CREDITCARDISSUER
• PREFIX
• LASTNAME
• AMOUNT
• DATE
• DOB
• COMPANYNAME
• BUILDINGNUMBER
• STREET
• SECONDARYADDRESS
• STATE
• EMAIL
• CITY
• CREDITCARDNUMBER
• SSN
• URL
• USERNAME
• PASSWORD
• COUNTY
• PIN
• MIDDLENAME
• IBAN
• GENDER
• AGE
• ZIPCODE
• SEX

Training hyperparameters

The following hyperparameters were used during training:

Hyperparameter	Value
Learning Rate	5e-5
Train Batch Size	16
Eval Batch Size	16
Number of Training Epochs	7
Weight Decay	0.01
Save Strategy	Epoch
Load Best Model at End	True
Metric for Best Model	F1
Push to Hub	True
Evaluation Strategy	Epoch
Early Stopping Patience	3

Training results

Epoch	Training Loss	Validation Loss	Precision (%)	Recall (%)	F1 Score (%)	Accuracy (%)
1	0.0443	0.038108	91.88	95.17	93.50	98.80
2	0.0318	0.035728	94.13	96.15	95.13	98.90
3	0.0209	0.032016	94.81	96.42	95.61	99.01
4	0.0154	0.040221	93.87	95.80	94.82	98.88
5	0.0084	0.048183	94.21	96.06	95.13	98.93
6	0.0037	0.052281	94.49	96.60	95.53	99.07

Author

[email protected]

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2