Real-time Data Privacy Protection With Adaptive ?-event Differential Privacy

With the development of Internet of Things,cloud computing,and mobile Internet,data in smart devices and applications are experiencing an unprecedented surge.These data are gathered from human-centric networks and device-based applications smart devices and applications.Using data mining and machine learning,we can extract information and mine relevance from these massive raw data to improve our insights.Yet,data mining introduces numerous security risks after convenience to human beings.Privacy preserving is the first to bear the brunt.In view of the increasing attention on personal privacy,data privacy protection has become a hot research topic.Numerous works presented a series of schemes to cope with privacy leakage.Most of these schemes are implemented based on traditional encryption algorithms.These may cause high computational complexity and are easy to be cracked.As a result,some studies introduced the differential privacy protection technology to preserve privacy.The typical differential privacy protection technologies only consider data accuracy,making it difficult to measure and evaluate the overall system security for streaming data.In addition,current differential privacy solutions still should further improve the practicability and the availability.In this thesis,we studies the application of differential privacy protection in streaming data.The main contributions are as follows.Firstly,to make the differential privacy protection algorithm better applicable to real-world scenarios,a QoP-based adaptive ?-event-level differential privacy algorithm is proposed.The proposed algorithm can adjust the window length of privacy protection in real time according to the changing state of data,and improve the availability and versatility of the algorithm.Secondly,the concept of privacy quality,QoP,is defined as the privacy protection metric.This metric takes into account both the error of the statistical data and the window length of the privacy protection.This new measurement method can evaluate the differential privacy protection algorithms more comprehensively,especially for streaming data scenarios.Thirdly,to avoid excessive noise introduced by the perturbation algorithm in differential privacy,an intelligent packet-based perturbation algorithm is proposed.The algorithm utilizes the machine learning technology to intelligently group data.It can effectively reduce errors introduced by the perturbation mechanism in the differential privacy algorithm,and achieve higher data precision under higher privacy requirements.Fourthly,for real-time data transmission in fog computing,the corresponding privacy protection solutions and application cases are given,which can provide strong privacy protection for distributed sensor data.This thesis consists of 5 chapters,including 11 figures,5 tables and 100 references.