Mining Multiple Private Databases using a Privacy Preserving kNN Classifier

Data mining technologies are popular for identifying interesting patterns and trends in large amounts of data. With the advent of high speed networks and easily available storage, many organizations are able to collect large amounts of data. On one hand, these organizations would like to mine their data to understand and discover interesting patterns; on the other hand, many legal and commercial reasons may prevent the organizations from sharing their data. In such situations, privacy preserving data mining tools become critical they allow data from many organizations to be mined securely and with a minimum of information disclosure. In this paper, we present a framework for mining multiple private databases using a privacy preserving k Nearest Neighbor (kNN) classifier. We develop a general model for privacy preserving kNN classification and present algorithms for realizing this model. We specify requirements that all privacy preserving classifiers should strive to achieve and analyze how well our algorithm achieves these requirements. This is the first paper to show how kNN classification can be achieved in a privacy preserving manner. A novel feature of our algorithm is that it offers a trade-off between accuracy, efficiency and privacy. Thus, it can be applied in a variety of problem settings and can meet different optimization criteria.