A SURVEY PAPER ON IMPROVED METHOD FOR PRIVACY PRESERVING DATA MINING CONSIDERING LINEAR AND NON- LINEAR ATTACK

Abstract

Privacy Preserving Data Mining (PPDM) is used to extract knowledge from dataset and a preserve the privacy before data can be release. The study of perturbation based PPDM approaches introduces random perturbation is the number of changes made in the original data. The limitation of previous solution is single level trust on data miners but new work is perturbation based PPDM to multilevel trust. When data owner sends number of pertubated copy to the trusted third party that time adversary cannot find the original copy from the pertubated copy means the adversary diverse from original Copy this is known as the diversity attack. To prevent diversity attack is main goal of MLT-PPDM services. Malicious data miners have access the different pertubated Copy of the same data through the various mean to combine this all the diverse copy to get original data very accurately this goal of with respect to privacy. In this work a user produce large number of pertubated copies of its data for random trust level on demand. Hence the user having maximum flexibility. The previous work is limitated only for linear attack. But proposed result is work on the non-linear attack also. In previous anonymization techniques generalization and bucketization related to the privacy of individual information to the adversary. The generalization involves loss of information and bucketization approach does not protection from membership disclosure, but in proposed approach slicing with tuple grouping algorithm partitioned from membership disclosure and also can handle large amount of data.
Key words: Diversity Attack, K-Anonymity, Multi-Level Trust, Non-Linear Attack, Parallel Generation. Introduction