Noise averaging effect of privacy-preserving data mining in time-series databases

초록

In recent years, privacy-preserving data mining (PPDM) has been extensively investigated. The aim of PPDM algorithms is to obtain accurate mining results while protecting at the same time sensitive information. In case of time-series data, random data perturbation, which adds random white noise to original time-series, is widely used for preserving the sensitive information. The random data perturbation, however, incurs a critical problem of decreasing accuracy of mining results since the noise severely distorts the original time-series. To solve this problem, we present a novel notion of noise averaging effect, which is derived from a simple intuition that the average of random noise converges to 0. We note that piecewise aggregate approximation (PAA) exploits the noise averaging effect, and it can alleviate the problem of decreasing accuracy of mining results. We apply the noise averaging effect to the basic random perturbation and the Wavelet-based perturbation and analyze their experimental results. ICIC International ©2011 ISSN.

키워드