DataSentinel : Advanced Detection of Sensitive Information with Machine Learning

Protecting sensitive information, such as passwords, IP addresses, email IDs, and private keys, is crucial for preventing data breaches and complying with regulatory standards. The paper introduces a new method that improves the detection of such information by combining Bidirectional Long Short-Term Memory (BiLSTM) and the Robustly Optimized BERT Approach (RoBERTa) models. These models are fine-tuned using a diverse dataset, enabling better contextual understanding and detection accuracy. The solution is implemented as a Windows Executable that can monitor and analyze various file formats, including PDFs, DOCX, and email texts. Identified sensitive information is securely stored using hash functions, ensuring strong data protection. This approach enhances data security and facilitates compliance with data protection regulations.