SP-SV: Balancing Privacy and Usability in Multi-Attribute Health Data Environments

Abstract

Ensuring data privacy has become increasingly vital for entities such as statistical bureaus, healthcare providers, and other data-dependent institutions. Although numerous data publishing techniques have been proposed, most are limited to managing datasets with a single sensitive attribute, rendering them ineffective in multi-sensitive environments. Conventional models like k-anonymity and ℓ-diversity face significant limitations in preserving privacy across multiple sensitive fields and often lack flexibility for tailored data handling. To address this gap, we propose the Sensitivity Preservation–Securing Value (SP-SV) Method, a novel framework that builds upon Differential Privacy to support anonymization across multiple sensitive attributes. SP-SV employs an adaptive noise injection strategy that dynamically adjusts according to the sensitivity level of the data, rather than relying on uniform noise distribution. This selective mechanism enhances protection for highly sensitive data points while retaining analytical value. Using synthetically generated datasets based on real-world healthcare records (comprising 6,000 entries), our experiments reveal that SP-SV maintains data utility with a maximum variation of only 12.45% for Sciatica and 12.50% for Fungal Infection. Compared to systems like Airavat, which apply static noise levels, SP-SV demonstrates superior flexibility and efficiency by aligning noise with data sensitivity.