A New Algorithm-independent Method for Privacy-Preserving Classifica- tion Based on Sample Generation

With the development of data mining technologies, privacy protection is becoming a challenge for data mining applications in many fields. To solve this problem, many PPDM (privacy-preserving data mining) methods have been proposed. One important type of PPDM method is based on data perturbation. Only part of the data-perturbation-based methods is algorithm-irrelevant, which are favorable because common data mining algorithms can be used directly. This paper proposes a new algorithm-irrelevant PPDM method for classification based on sample generation. This method is a data-perturbation-based method and has three steps. First, it trains classifiers use the original data. Then, it generates new samples as the perturbed data randomly. Finally, it use the classifiers trained in the first step to predict these samples' category. The experiments show that this new method can produce usable data while protecting privacy well.