Fit different types of models to loss given default (LGD) data. This example demonstrates the following approaches:

Work with consumer (retail) credit panel data to visualize observed probabilities of default (PDs) at different levels. It also shows how to fit a Cox proportional hazards (PH) model, also known as Cox regression, to predict PDs. In addition, it shows how to perform a stress-testing analysis, how to model lifetime PDs, and how to calculate the lifetime expected credit loss (ECL) value using portfolioECL.

Work with consumer credit panel data to create through-the-cycle (TTC) and point-in-time (PIT) models and compare their respective probabilities of default (PD).

Create, train, and compare three deep learning networks for predicting credit default probability.

Train a credit risk for probability of default (PD) prediction using a deep neural network. The example also shows how to use the locally interpretable model-agnostic explanations (LIME) and Shapley values interpretability techniques to understand the predictions of the model. In addition, the example analyzes model predictions for out-of-sample values and performs a stress-testing analysis.

Calculate and use data and model metrics to investigate the biases that exist in a model.

Use bias mitigation with a credit scorecard model to make it more fair.

Use disparate impact removal as a pre-processing technique in bias mitigation to a credit scorecard model to reduce bias in the model.

Different techniques for interpreting and explaining the logic behind credit scoring predictions.