Study On Microaggregation Algorithms For Privacy Protection

In modern society, the rapid development of network enables more and more data to be shared. The growing information has brought us great convenience in our daily life and work. However, the user's private information often leaks in the process of releasing microdata. Publishing data about individuals without revealing sensitive information about them becomes an important problem. K-anonymization is an important technique during microdata publication, it can simplicity and practicability protect private information. Recently, microaggregation technique has been introduced to combine with k-Anonymization in order to get better performance. The paper studys the microggregation techniques for Privacy Protection. The main tasks of this paper are listed as followings:The paper studies the existing technique for privacy protection, especially the k-Anonymity and l-diversity modes. By comparing the algorithms of k-Anonymity, the paper make a conclusion about the advantages and disadvantages of the algorithms. At last the paper proposes some effective solutions based on the problems.Then the paper goes deep into the microaggregation technique, analyzes the assessment model of the algorithms. The problem is that the runtime of microaggregation algorithms will be long when large data is processed. According to the problem, the paper proposes a method to be more effeicient for the algorithms. By grouping nearest records into one cluster first, the runtime can be reduced.And then, the paper analyzes the disadvantages of the clustering for sensitive attribute. Although the clustering can reduce runtime of the algorithms, it will cause a problem about senstive attributes. To solve the problem, the paper proposes an effeicient microaggregation algorithm—MKL algorithm. When clustering the records, we should keep the distribute of senstive attributes in each cluster unchanged. Then for each cluster, we get k nearest records into one group which has at least l distinct sensitive values. so the anonymity table satisfies l-diversity constraint which can resist homogeneity attack and background knowledge attack.At last ,the paper propose a method to decide the value of m,which can make the algorithm more practical.Finally, the algorithm is implemented with Adult database from machine learning center of University of California, and the paper analyzes results of the experiments. By comparing the algorithms on time cost, information loss and privacy disclosure risk, the paper make an assessment of the algorithms. The experimental results show that MKL algorithm can make anonymity table satisfy l-diversity constraint and need less runtime, it can get better performance than other algorithms.