A study of privacy preservation and classification approaches in data mining applications

Abstract

Privacy preserving data mining is a field to protect the privacy of newlinesensitive data and also provides a valid data mining results. The data newlineperturbation techniques are the well-liked models which perform the data newlinetransformation process before publishing data to the data miners. There is a newlinenecessity to prevent diversity attack by adequately correlating perturbation newlineacross copies at different trust levels in an organization. To achieve high newlineprivacy guarantee and zero- loss of accuracy, various perturbation techniques newlineare used for different classifiers. By removing irrelevant and redundant newlinefeatures from the dataset, the performance of the classifiers can be improved. newlineIn the initial stage of research, a Hybrid Gaussian Noise Distribution newline(HGND) perturbation method is addressed for maintaining sensitive data newlineamong multiple privacy level. The perturbed data generation process is done newlinein three different ways such as parallel generation, sequential generation and newlineon-demand generation for all the additive, multiplicative and hybrid newlineperturbation methods. A data owner can produce perturbed copies through on demand newlinebasis with respect to privacy levels. Higher privacy level data miner newlinecan access only less perturbed data. We proved that our model produces best newlineresults against diversity attacks, in which attacker may access the collection of newlinethe perturbed copies. But our model prevents them from jointly reconstructing newlinethe original data more accurately newline newline