Estimating Kriging-based predictions with privacy

Abstract

Kriging is a well-known prediction method. It interpolates the value of an unmeasured location from nearby measured locations. In a traditional Kriging interpolation, a client (an entity that is looking for a prediction for a specific location) asks help from a server (an entity that holds enough measurements collected for Kriging interpolations in a region). Predictions are estimated base on location data and measurements, which are considered confidential data. Neither the client nor the server wants to reveal their private data to each other. Although Kriging is increasingly becoming popular and widely used for estimating predictions, it fails to protect confidentiality. Thus, clients and servers might hesitate to participate in Kriging interpolations. In this study, we investigate how to provide Kriging-based predictions without violating data owners' privacy. We propose a scheme, which helps the clients and the servers perform Kriging interpolations while protecting their confidentiality. In other words, our method does not allow them from deriving information about each other's private data. We show that the proposed scheme protects privacy and it does not cause any accuracy losses. We also analyze it with respect to inevitable additional costs, which do not affect online performance. Our analyses show that the proposed scheme is able to provide accurate predictions efficiently while preserving privacy