Abstract:
Arti cial intelligence is applied in more and more high impact elds, such as the
banking world. In this elds it is often not acceptable to only give predictions
and no explanation. Therefore the eld of Explainable Arti cial Intelligence
is also growing. In this thesis a random forest model for money laundering
detection for the Volksbank was developed to replace the current rule-based
model. This resulted in a large drop in false positives. Two explainable models,
SHAP and LIME, were placed over the random forest model and tested on
reliability and user friendliness using expert reviews. The SHAP model was
preferred over the LIME model.