Optimizing the Trade-off Between Classification Accuracy and Data Privacy in the Area of Data Stream Mining

Abstract

Data perturbation has grabbed the attention of data mining, as preserving the privacy of the data is crucial, especially in sensitive data. But the perturbation process negatively affects the accuracy of predictions, generating a trade-off between privacy and accuracy. We propose seven
different cumulative noise addition based perturbation methods combining a set of techniques such as logistic function, use of absolute noise values, and cycle-wise noise addition as possible solutions for this accuracy-privacy trade-off issue. These techniques are introduced to optimize the trade-off between classification accuracy and data privacy by controlling the maximum noise level.Moreover, we evaluate the performance of the proposed methods compared to the state of art of the noise addition-based perturbation methods to select the best of them.