Privacy-Preserving Kriging Interpolation on Distributed Data

Abstract

Kriging is one of the most preferred geostatistical methods in many engineering fields. Basically, it creates a model using statistical properties of all measured points in the region, where a prediction value is sought. The accuracy of the kriging model depends on the total number of measured points. Acquiring sufficient number of measurement requires so much time and budget. In some scenarios, private or governmental institutions may collect geostatistical data for the same or neighbor region. Collaboration of such organizations may build better models, if they join their data sets. However, due to financial and privacy reasons, they might hesitate to collaborate. In this study, we propose a solution to build kriging model using distributed data while preserving privacy of each data owners and the client that requests prediction. The proposed scheme creates a kriging model on joint data of all parties who wants to collaborate. We analyze our solution with respect to privacy, performance, and accuracy. Our solution has extra costs; however, they are not that critical. We conduct experiments on real data sets to show that our scheme gives better result than the model created on insufficient measured data.