Discord Region Based Analysis to Improve Data Utility of Privately Published Time Series

Privacy preserving data publishing is one of the most important issues of privacy preserving data mining, but the problem of privately publishing time series data has not received enough attention. Random perturbation is an efficient method of privately publishing data. Random noise addition introduces uncertainty into published data, increasing the difficult of conjecturing the original values. The existing Gaussian white noise addition distributes the same amount of noise to every single attribute of each series, incurring the great decrease of data utility for classification purpose. Through analyzing the different impact of local regions on overall classification pattern, we formally define the concept of discord region which strongly influences the classification performance. We perturb original series differentially according to their position, whether in a discord region, to improve classification utility of published data. The experimental results on real and synthetic data verify the effectiveness of our proposed methods.