Privacy preserving classification based on randomization and reconstruction

Privacy preserving classification is to develop a classifier without precise access to the original data. In order to improve the applicability with higher privacy and better accuracy, we present a novel Privacy Preserving Naive Bayes (PPNB) classification method that consists of two steps: first, the original data set is distorted by a new. randomization approach; second, a naive Bayes classifier is implemented on the distorted data set to predict the class labels for unknown samples. Besides being analyzed in applicability, privacy, accuracy, and efficiency, the effectiveness of our PPNB classification method is also validated by the experiments.