Differential Privacy Releasing Of Healthcare Streaming Data

With the popularity of wearable medical or health devices,healthcare big data mining and analysis has become a research hotspot.Healthcare streaming data plays an important role in disease monitoring and prevention.However,the healthcare streaming data contains the patients' sensitive information.If we released the data directly,and attackers performed background knowledge attack,the patients' privacy information could be easily obtained.We use the ?-event privacy to protect the privacy of healthcare streaming data.Direct application of the existing ?-event privacy algorithm cannot meet the requirements of the availability of healthcare streaming data.Meanwhile,there are two obvious characteristics of healthcare streaming data: fluctuation over time and multi-events fluctuation.Therefore,we improved the privacy algorithm of the ?-event privacy according to these two features to meet the requirements of the availability of the healthcare streaming data.At different timestamps,the counting statistics of healthcare streaming data fluctuate greatly.Existing releasing algorithm BA(Budget Absorption)for streaming data did not take into account the fluctuation of streaming data.In order to reduce error,we introduced ? based on BA.? is the weight of dissimilarity which is the mean absolute distance between statistic and latest non-null released value when comparing with the theory of noise,and ? can be adjusted adaptively.The variance in the sliding window was calculated to judge whether the fluctuation of flow data was stable.When the fluctuation of stream data is stable,we increased the value of ? to reduce the number of releasing timestamps,and increase the proportion of privacy budget absorption.When streaming data is volatile,we reduced the value of the ? to increase the number of release points.AdaptiveBA(Adaptive Budget Absorption),a privacy releasing algorithm based on Adaptive dissimilarity parameters was designed.The algorithm adapts to the fluctuation of healthcare streaming data over time and reduces the error of privacy releasing.In the healthcare multi-event streaming data,the count statistics' fluctuation features of each event are different.If the count statistics of few events changed and other events basically are stabled,then only few events affect the dissimilarity of the existing privacy publishing algorithm.In this case,the value of the stable events can be approximated by the historical released value.Using offline or online methods calculated the variance of each event within the time window.By comparing the variance and the given threshold,the events were divided into two groups.A privacy releasing algorithm EventGroupBA(Event Group Budget Absorption)was designed.According to the fluctuation features of each event,the algorithm adjusted the releasing frequency of each event to reduce the noise added in the publishing process,and reduce the error of privacy publishing.Experiments were carried out on several real healthcare streaming data sets,and the results showed that under the same privacy budget,AdaptiveBA algorithm could reduce the errors in the privacy releasing of single event stream and multiple event stream at the same time,against BA algorithm.EventGroupBA algorithm optimizes BA algorithm and can effectively reduce the error of multi-event stream privacy releasing under the same privacy budget.