The article GDA Score Overview describes how GDA utility and defense scores are derived. For those of you who don’t need so much detail, this article gives a brief summary of how to read the GDA Score graphics.
Every GDA score consists of a several individual score components, shown on the right, and a combined score shown on the left. The combined score is composite derived from the individual components.
All of the bars are oriented so that higher is better: either better utility or better defense (stronger anonymity). This way one can at-a-glance see which score components are positive and which are negative. The numeric value associated with each individual score is also shown. Note that a higher bar does not always mean a higher number.
The combined score is color-coded with five colors ranging from very good to terrible. The individual components are not color coded.
The color coding, as well as the orientation of the bars, allows one to get a quick overview of an anonymization methods utility and defense across different dimensions (different datasets, different attacks, etc.)
For many of you, this quick overview may be all you need. For a slightly deeper understanding, the following figure gives a brief interpretation of each individual component for the defense score. The GDA Score Overview article explains these in more detail.
Note that the most important components are the confidence improvement and the probability. An anonymization method that gives either low confidence improvement or very low probability of high confidence improvement can be regarded as strong.
The utility score is a little simpler. It has two components, accuracy and coverage. Coverage shows how much of the dataset can be observed, and accuracy shows how much the observable data is distorted.
