This invention provides a localized
differential privacy protection method for medical data. It employs a perturbation
algorithm to perturb the data, with the privacy budget set as follows: sensitivity levels are applied to the relevant
disease types; based on the sensitivity level, a corresponding privacy budget is set for each
disease, where higher sensitivity levels correspond to smaller privacy budget values. This invention assesses
disease sensitivity from multiple dimensions, including
genetic risk, social discrimination, and
treatment costs, achieving differentiated dynamic privacy
budget allocation. Smaller values ​​are allocated to
highly sensitive diseases to enhance
privacy protection, while larger privacy budget values ​​are allocated to less sensitive diseases to reduce data
distortion. This solves the problem of insufficient targeted protection caused by fixed privacy budgets in existing methods. Furthermore, the accuracy against background knowledge attacks is significantly lower than 1.2 times that of random guessing, and the
privacy protection effect meets the stringent requirements of the formal definition of localized
differential privacy.