An efficient perturbation approach for multivariate data in sensitive and reliable data mining

Due to the rapid enhancement of technology, cloud data is increasing rapidly which contains individuals' sensitive information such as medical diagnostics reports. While extracting knowledge from those sensitive data, both privacy of individuals' and the utility of data should be preserved which is a crucial concern in data mining related activities. Though therein exist several methods to preserve privacy, a single method can not maintain the harmony interim privacy and data utility. Often achieving individuals' privacy leads to the loss of the data utility and the opposite is true also. To address the vital issue, a four-stage data perturbation approach, called NRoReM, is proposed in this work based on normalization, geometric rotation, linear regression, and scalar multiplication for sensitive data mining. The proposed approach is experimented with over ten UCI data set using three benchmark classifiers. The empirical exploration of privacy protection, attack resistance, information entropy analysis, data utility, and error analysis exhibits that NRoReM preserves both privacy of individuals' and data utility on a larger scale for 90% of the data set than 3-Dimensional Rotation Transformation (3DRT) and 2-Dimensional Rotation Transformation (2DRT).