|
background - solutions
Various solutions have been proposed in the literature to reconcile the risks of location privacy with the requirements of public health practice. Among them have been suggestions for increased access control, suppression of information, aggregation, k-anonymisation methods, the use of software agents and mathematical transformations.
Among the latter, a method referred to as random perturbation has been the subject of various studies. In a classical random perturbation, a given point is randomly moved within a certain distance to mask it's original location. Of course this has to be done with consideration to where the point falls. For example, if it happens to be in a sparsely populated rural area, then it'll have to be randomly displaced - or perturbed - a much larger distance to achieve anonymity than a point that occurs in a densely populated urban area. In the more recent studies, the underlying population is adjusted for by applying weights to the perturbation distance. The perturbation must also ensure that the point doesn't fall in an implausible or improbable location - for example, a water body - and another identified issue is the fact that if the point is randomly re-positioned, then performing the perturbation many times and taking the mean of the results will approximate the original location.
To date, no solution has been for public health in terms of such a transform that incorporates location-privacy right in with the transformation of the rest of the data - in essence, a "privacy by design" approach which includes location privacy. The proposed transform is the first one of its kind in public health, acting on multiple dimensions will giving the user maximum flexibility and control over the results. The methodology has been described in the upcoming paper published ahead of print in Methods of Information in Medicine - click on Multidimensional Point Transform (also under Publications to the right) to access the article.
A static image showing what a tool interface may look like has been designed and can be accessed here. |
