The Impact of Linear Transformation on the Effectiveness and Security of the Privacy Preserving Data Mining Process

Abstract

Since data mining techniques can extract useful knowledge from data more and more people are devoted to this field The data for mining generally contain personal records and hence people pay more attention on preventing their private data from being disclosed This study attempts to establish a procedure to ensure the effectiveness and security of the original data Data are transformed by piecewise linear functions before sending to data analysts who will apply classification methods on the transformed data Transmitting data are also protected by perturbation and encryption processes The classification models produced by data analysts can be sent back to data providers who will restore the models for the data analysts This restoring process is designed to ensure that data analysts can have models for classifying new instances According to the experimental results on ten data sets the more pieces a linearly function has the higher security can be achieved for algorithms decision tree and rule-based classifier The data analyzed by algorithms logistic regression and support vector machine should not be transformed by multi-piece linear functions because the accuracies of the original and transformed data resulting from these two algorithms will be different

Date of Award	2020
Original language	English
Supervisor	Tzu-Tsung Wong (Supervisor)