Statistical information recovery from multivariate noise-multiplied data, a computational approach

Authors

Yan-Xia Lin, University of WollongongFollow
Luke Mazur, University of WollongongFollow
Rathin Sarathy, Oklahoma State University
Krishnamurty Muralidhar, University of Oklahoma

RIS ID

127143

Publication Details

Lin, Y., Mazur, L., Sarathy, R. & Muralidhar, K. (2018). Statistical information recovery from multivariate noise-multiplied data, a computational approach. Transactions on Data Privacy, 11 (1), 23-45.

Abstract

This paper proposes a computational statistical method for multivariate confidential numerical microdata. The method can be employed for recovering some commonly interesting statistical information present in the microdata from noise-multiplied data. Estimating the parameters in linear regression without using the original data directly becomes feasible. This paper demonstrates that some statistical information can be recovered reasonably well for certain types of original data while the level of disclosure risk is under control if the multiplicative noises used to mask the data are appropriate. This paper presents an alternative approach for sharing the statistical information of multivariate confidential data and carrying out data mining with multidimensional sensitive data, an area of growing interest. An R package MaskJointDensity is built for implementing the method 1 .