Privacy Preserving Classification in Vertically Distributed Datasets

Nowadays, preserving the privacy of data is an important issue during data mining techniques. several algorithms have been proposed to preserve the privacy of data. The most important problems with these algorithms are their unprovable privacy, the low importance of considering the adversary’s background knowledge and lack of dimensionality reduction process over the original data. In this paper, differential privacy mechanism has been used to prove the privacy of vertical distributed data to be used for classification. In differential privacy, it is no matter what background knowledge has an adversary about the published data. Also, Haar wavelet transform has been used for dimension reduction of the original data.The privacy of data has been proved mathematically and the accuracy of data measured using the K-NN algorithm. Finally, it has been mathematically proved that the proposed algorithm adds noise with less standard deviation to the data than the compared algorithm, resulting in higher classification accuracy. The result shows that our algorithm has more secure compared to previous algorithms.