Deep dive

Home - - Deep dive

Details of the technical approach to develop the credit scoring
model based on PSD2 data

• Created over 3000 features, i.e. composite variables
generated from the raw transaction data

• 20 different scenario’s evaluated in
detail, many more while analysing data
and validating different approaches

• For each PSD2scenario, we trained 40
models with different test/train set splits to
determine stability of errors (bootstrapping)

• Developed several hundreds of codes, routines
and visualisations in Tableau, Python and R

• 6 different machine learning techniques
tested, including logistic, linear regression,
gradient boosting and neural networks