We try to solve discrimination on the German Credit Data Set used in Verma & Rubin (2018) with the model presented in Kilbertus et al. (2017).

The first thing we try is analysing the data and try to find a useful causal model to apply the method. We build a Classifier implementing the idea from Kilbertus et al. (2017). Since it is very difficult to set up a causal model, we restrict our model to the following features:

• credit duration
• credit history
• credit amount
• present employement since (proxy variable)
• sex
• age (protected attribute)
• job
• foreign worker

While it is for sure not a complete model it should be able to show if the idea the paper presents is working.

The first thing we have to do is to build a causal model graph.

The second thing we have to do is to reduce the dataset down to the needed features.

Then we can start applying the proposal from Kilbertus et al. (2017).