Stable Feature Selection with Privacy Preserving Data Mining Algorithm

Abstract

Data mining extracts previously not known and valuable type of patterns and information procured from large storage of data that is archived. In the last few decades, the advancements in internet technologies results in enormous increase in the dimensionality of the dataset concerned with data mining. Feature selection is an important dimensionality reduction technique as it improves accuracy, efficiency and model interpretability of data mining algorithms. Selection of feature and its stability may be perceived to be the robustness of the algorithm for feature selection which helps selecting similar or the same subset of features for small perturbations in the dataset. The essential purpose of data mining that is used for the preservation of privacy is the modification of original datasets by means of a method to preserve privacy of the individuals and work out subsequent data mining algorithm to get information from it. This perturbation of the dataset will affect the feature selection stability. There will be a correlation between privacy preserving data mining and feature selection stability. This paper explores on this problem and also introduces a privacy preserving algorithm which has less impact on feature selection stability as well as accuracy.