A Machine-Learning-Based Tool for Redaction, Masking and Anonymization of Sensitive Data

Abstract: Sensitive information appears in many forms of digital data, including documents, text records, images, logs and other files. Traditional redaction approaches often depend on manually defined rules, which can be difficult to maintain across diverse input formats and may not provide a controllable level of anonymization. RE-DACT is proposed as a natural-language-processing and machine-learning-based redaction tool designed to redact, mask or anonymize sensitive information while preserving the structural and logical usefulness of the original data.

The proposed system accepts user-defined levels of redaction and aims to provide customized output. A learning component can be introduced to improve its ability to generate realistic synthetic replacements while maintaining the required data characteristics. The research design emphasizes security, minimal data retention, support for multiple input sources, computational efficiency, usability and operation in both web-based and offline environments. Evaluation is proposed using precision, recall and F1-score on an open-source testing dataset, together with measurements of speed, computing usage, preservation of structure and effectiveness of the selected redaction level.

Keywords: Data Redaction, Anonymization, Masking, NLP, Machine Learning, Synthetic Data, Privacy, Cybersecurity, Data Security