The Effects of Privacy Preserving Data Publishing based on Overlapped Slicing on Feature Selection Stability and Accuracy

Abstract

Feature selection is vital for data mining as each organization gathers a colossal measure of high dimensional microdata. Among significant standards of the algorithms for feature selection, the primary one which is currently considered as significant is feature selection stability along with accuracy. Privacy preserving data publishing methods with various delicate traits are analyzed to lessen the likelihood of adversaries to figure the touchy values. By and large, protecting the delicate values is typically accomplished by anonymizing data by utilizing generalization and suppression methods which may bring about information loss. Strategies other than generalization and suppression are investigated to diminish information loss. Privacy preserving data publishing with the overlapped slicing technique with various delicate ascribes tackles the issues in microdata with numerous touchy attributes. Feature selection stability is a vital criterion of data mining technique because of the accumulation of ever increasing dimensionality of microdata due to everyday activities on the World Wide Web. Feature selection stability is directly correlated with data utility. Feature selection stability is data centric and hence modifications of a dataset for privacy preservation affects feature selection stability along with data utility. As feature selection stability is data-driven, the impacts of privacy preserving data publishing based on overlapped slicing on feature selection stability and accuracy is investigated in this paper.