Personal Identity Information Detection using Synthetic Dataset

Safeguarding Personal Identity Information (PII) is essential to protect privacy, prevent identity theft, comply with laws, maintain trust, and ensure national security. Unauthorized access or disclosure of PII can have severe consequences, including financial fraud and reputation damage. With the increasing attention paid to safeguarding personal details online, the identification of PII has become crucial for protecting sensitive and private information. This paper addresses the effectiveness of PII detection using different Natural Language Processing (NLP) techniques trained with synthetic PII data. Three different methods were experimented with for generating appropriate synthetic data. Furthermore, the data is pre-processed to ensure that the final generated dataset is meaningful and usable. This generated dataset will be used to train state-of-the-art NLP models, namely spaCy, BERT, and biLSTM. It is important to note that the synthetic data generated will be processed based on each model’s required format and then trained or fine-tuned. Performance evaluation is conducted for each model, where the best-performing model will be used to create a sustainable PII detection tool. The metrics precision, recall, and F1 score are used to compare the performance of the above NLP models.