Predictive delimiter for multiple sensitive attribute publishing

Mining medical records to extract useful information is bound to exposure of individual personal data. Typical Electronic medical record houses multiple sensitive symptom attributes of patients. Dealing with multiple categorical sensitive attributes is a challenge in terms of managing privacy of individual. Objective of this research aims in retaining the privacy while delivering accuracy during mining. Background work on K-anonymity and L-diversity models reveal failure or reduction in accuracy when handling multiple sensitive attributes. It is easy to predict the individual's disease by understanding the symptoms associated to it. Investigating the symptoms and their relationship can harness the privacy of the individual. Proposed predictive delimiter algorithm targets the hidden knowledge in the combination of symptoms. This method redistributes the sensitive symptoms such that the data intruder is unable to predict the disease associated with the patient. Experimental results with privacy yield of 99.78% and accuracy yield of 62.3% validates the efficiency of the proposed algorithm.