Figure 3 - Flowchart of the methodology applied
In the former methodology, clustering techniques were applied to find
peer groups using a small subset of features from the same initial pool,
and in the first approach after having applied the linear and nonlinear
models , the score was calculated inside every cluster; in the second
approach the predictive models were applied in every cluster to later
assigned the score within the cluster as well.
As at the end, the features used were the same in both cases, it was
prove that it was not only not helping but distorting the fairness.
Moreover, using clustering techniques was a method to get a better
performance of the predictive method but at the end it doesn’t improve.
However, it is still interesting to discuss the process that it can be
implemented in a future work.