On Feature Selection Stability and Privacy Preserving Data Mining: A Data Perspective

Abstract

The data relating to data mining has turned out to be profoundly multi-dimensional in the recent past. It is also to be noted that such dimensionality has rapidly expanded over time. Moreover, in light of the positive assessment norms for enhanced data mining concerts, feature selection opts for a petite subset of the significant features from the original dataset. The stability of the feature selection is a key criterion in feature selection algorithms. Moreover, the most important aspect is its sturdiness in reducing the disturbances in the training data or in the expansion of the most recent samples. Lately, it has been demonstrated that the stability of the feature selection usually centers on data, and that it is not entirely unbiased in terms of algorithm. The privacypreserving data mining changes a portion of the sensitive and quasi-identifying attributes in order to keep the conceivable re-identification of an individual’s tuple through intrusive or malignant data miner and brings a choppy privacy conserved dataset. Since the stability of the feature selection relies primarily upon data, the stability of the feature selection lessens with such privacy-preserved choppy datasets. Besides, the privacy preserving ruffling associates stresses the stability of the selection of features and data utility. Picking proper privacy-preserving data mining technique with significant privacy-preserving ruffling to enhance feature selection stability alongside the greater privacypreservation and data utility is consequently a challenging issue in the field of research. Hence, the present paper intends to highlight the issue with reference to the three algorithms for privacy-preserving data mining and their relative analysis. © 2020, World Academy of Research in Science and Engineering. All rights reserved.