Adaptive Voronoi Masking. A method to protect confidential discrete spatial data

Abstract

Geomasks have been developed to assure the protection of individuals in a discrete spatial point data set by transferring the data points to a new location. Several obfuscating techniques exist but the risk of false re-identification is a commonly discussed problem. This thesis develops an alternative approach, referred to as Adaptive Voronoi Masking (AVM), which is based on the concepts of Adaptive Aerial Elimination (AAE) and Voronoi Masking (VM). It considers the underlying population density by establishing areas of K-anonymity in which Voronoi polygons are created. Complementary to many other geomasks, the proposed method considers the underlying topography and displaces data points to street intersections decreasing the risk of false-identification immensely since residences are not endowed with a data point. The spatial characteristics of the new method are examined by the mean and median centers, the Nearest Neighbor Hierarchical Cluster Analysis (NNHCA), Ripley’s K-function, a visualization of the masked data points, and are subsequently compared with the output of AAE, VM, and Donut Masking (DM). VM attains the best efficiency for the mean centers whereas DM does for the median centers. Regarding the NNHCA, DM demonstrates the strongest performance since it’s cluster ellipsoids are the most similar to those of the ODP in terms of orientation, size, number of clusters, and mean points. In regard to the Ripley’s K-function, AVM and DM succeed the other geomasks since they achieve the most similar values as the ODP retaining the point pattern of the original data sets. However, AVM clearly outperforms all methods regarding the risk of false re-identification as observed from the visualization of the MDP because no data point is moved to a residence which is a major step forward in academic research in geomasking techniques. Furthermore, AVM has the ability to maintain the spatial K-anonymity which is also done by AAE and partly by DM. Hence, based on these three factors, AVM is the best obfuscating method. Additionally, this research analyzes whether any rules and regulations exist to protect individual data. Hence, the European GDPR was investigated to examine whether personal - particularly locational and health data - are protected. Thereby, it is concluded that the GDPR is vaguely worded and not efficient enough to protect the individual in terms of data processing. Besides, most geomasks do not comply with the rules and regulations of the GDPR since their masked data points can be traced back to their original location disclosing an individual by an address. Unlike this, AVM complies with the rules and regulations since no data point is moved to a residence that might be associated with an address and, therefore, with an individual.