Frequent Symptom Sets Identification from Uncertain Medical Data in Differentially Private Way

Abstract

Data mining techniques are applied to identify hidden patterns in large amounts of patient data. These patterns can assist physicians in making more accurate diagnosis. For different physical conditions of patients, the same physiological index corresponds to a different symptom association probability for each patient. Data mining technologies based on certain data cannot be directly applied to these patients' data. Patient data are sensitive data. An adversary with sufficient background information can make use of the patterns mined from uncertain medical data to obtain the sensitive information of patients. In this paper, a new algorithm is presented to determine the top K most frequent itemsets from uncertain medical data and to protect data privacy. Based on traditional algorithms for mining frequent itemsets from uncertain data, our algorithm applies sparse vector algorithm and the Laplace mechanism to ensure differential privacy for the top K most frequent itemsets for uncertain medical data and the expected supports of these frequent itemsets. We prove that our algorithm can guarantee differential privacy in theory. Moreover, we carry out experiments with four real-world scenario datasets and two synthetic datasets. The experimental results demonstrate the performance of our algorithm.