Research Of Privacy Preserving Data Mining Based On Perturbation

In recent years, with the development and the improvement of databasetechnology, network technology and computer data storage capacity, data mining as apowerful data analysis tools, made great contributions in many fields and has wideapplication prospect. The emergence of increasingly data mining algorithms makespeople easily obtain more and more information from the social organizations.Therefore the privacy protection between individual, enterprise or institution isincreasingly important. At present, the most common way of privacy protection in datamining is data perturbation method which is based on the statistics, data partition andthe disturbance under association rules, and compared with traditional method, it is avery efficient and can better protect the privacy of personal data.Based on the thought of data perturbation for the privacy protection issues in datamining, a reasonable solution to ensure the privacy of users demand is discussed in thisthesis. The method of decision tree, partition environment and association rules arecarried on and researched.The main contents researched in this thesis are as follows:(1) Classification of privacy protection mining based on perturbation method is putforward. Data mining technology for the current privacy protection system isclassified. The privacy protection algorithms based on perturbation method underthe basic idea and principle are summarized. And finally, the analysis and assessingprinciples of the privacy protection algorithms from aspects of the difficulty andthe practicability are discussed.(2) On the basis of the research on the decision tree method, the structurecharacteristics of decision tree and disturbance algorithms are discussed. Thecombination of these two basic methods is researched to build new perturbationmethods in view of the decision tree structure properties. Farther more, the methodof decision tree with degradation idea is also researched. Under the ideas of thesetwo methods, two examples are shown to demonstrate the validity of them. (3) Through the original database partition method, combined with the ideas oforiginal data disturbance for privacy protection, the partition perturbation methodbased on the disturbance tree is discussed and a risk assessment to this method isanalyzed.(4) Based on association rules theory, the method of privacy protection of informationrules in DM is studied and two perturbation algorithms based on association rulesare discussed. Finally, a preliminary research to the privacy protection applicationof perturbation methods with association rules is shown.