Comparison of Spatial Clustering Techniques for Location Privacy

Abstract

Location privacy was born to deal with protection privacy issues which came with the massification of georeferenced data due to the frequent use of phones, social media, GPS services and other applications. This georeferenced data can be directly connected to users' personal information like religion, health and tracking, and can be used for different purposes, such as local analysis or selling it to third party companies, which represents a risk for individuals when the information is published or robbed without any protection through a location privacy protection mechanism - LPPMs. Many LPPMs have been proposed in different papers, one of them is called VoKA, a K-Aggregation offline technique. The methodology explained in this paper takes the first part of VoKA, a gridification process, and then applies two different spatial clustering algorithms, K-Means and DBSCAN, in order to protect each point of a dataset. To explain how this mechanism works, a dataset of Dengue registers in Barranquilla-Colombia and its outskirts was used, taking into account that this kind of data is considered sensitive. The results explain how this dataset can fit better with one of the algorithms and its respective metrics using squared error, point loss and heatmap comparisons.