Privacy-Preserving Data Mining Techniques for Sensitive Datasets

Abstract

The appraisal of Privacy-Preserving Data Mining (PPDM) has become a crucial research area in current times as a result of the incredible increase in the applications of data-driven applications that use sensitive data (health records, financial transactions, social networks, and governmental databases). Although the data mining techniques have been offering effective tools in the extraction of valuable knowledge, they facilitate great risks to personal privacy when they are applied to sensitive data. Unauthorized disclosure, inference attack, and breach of data has brought up serious ethical, legal, and regulatory issues. As a result, it is difficult to find the compromise between data utility and privacy protection. This essay outlines an extensive analysis of privacy ensuring data mining methods that allow secure privacy of sensitive data without compromising on the analysis accuracy. The paper systematically investigates the ways of anonymization, perturbation, cryptography, and hybrid privacy models. Besides that, newer privacy models include differential privacy, federated learning, and secure multi-party computation are discussed. A systematic approach is given to assess PPDM methods using privacy strength, data utility, computational complexity and scalability. The paper also reports on the findings of the experiments by comparing them, thus showing trade-offs between privacy and performance. The problems, problems under open research and direction are also discovered. The results highlight the lack of universal best practices because no single method is universally the best and the use of PPDM methods should be applied based on the application. The paper is intended to be a reference book of researchers and practitioners looking to have strong privacy preservation solutions such in sensitive data mining tasks.