Differentially Private Spatial Decompositions for Geospatial Point Data

Technological advancements in data analysis and data releasing have put forward higher security requirements, such as privacy guarantee and strictly provable security, this new area of research is called differential privacy. As for geospatial point data, the exiting methods use the tree structure to split the data space to enhance the data utility and usually adopt uniform budgeting method. Different from this, we propose a novel non-uniform allocation scheme for privacy budget which is a parameter to specify the degree of privacy guarantee. Firstly, the spatial data is indexed by quadtree, then, different privacy budget is allocated to each layer of quadtree using Fibonacci series features, and we designate this budgeting method as Fibonacci allocation. Experimental results show that Fibonacci allocation is significantly more accurate in data queries than the state-of-the-art methods under the same privacy guarantee level and fits for arbitrary range queries. Furthermore, data utility can be improved by post-processing and threshold determination.