抽象的な

Accumulative Privacy Preserving Data Mining Using Gaussian Noise Data Perturbation at Multi Level Trust

G.Mareeswari, V.Anusuya

Generally Data Mining develops the exact models about the collected data. Data perturbation, a widely employed and accepted Privacy Preserving Data Mining (PPDM) approach add random noise to original data , that prevent data miner to publish the accurate information about original data that is not allowed by data owner. Under the single level trust a data owner generate only one perturbed copy of its data with affixed amount of uncertainty. In this Project, the aim is to enlarge the scope of perturbation-based PPDM to Multilevel Trust (MLT-PPDM). In this system, different perturbed copies of same data are available to data miner at different trust level. If data miner is more trusted means, it can access the minor perturbed copy of the data. In case of malevolent data miner, may have access to differently perturbed copies of the same data and may combine these different copies to collaboratively induce more information about the original data that the data owner does not aim to release; this is the “DIVERSITY ATTACK”. Inhibiting such diversity attacks is the major provocation of providing MLT-PPDM services. In this project, the scope is to provide the additive perturbation approach where random Gaussian noise is added to the original data with arbitrary distribution, so the data miner will have no diversity gain and provide a systematic solution. This solution allows a data owner to generate perturbed copies of its data on demand at arbitrary trust levels.