15 June - Andreas Bueff Link https://eu.bbcollab.com/guest/97e5b2666aa74b30870a4c83ad03760a Speaker Andreas Bueff Title Interpretable Machine Learning for Stress Testing in Credit Risk Abstract To boost the application of machine learning (ML) techniques for credit scoring models, the blackbox problem should be addressed. We propose a measure based on counterfactuals to evaluate globally the interpretability of a ML credit scoring technique. Counterfactuals assist with understanding the model with regard to classification decision boundaries. The second contribution is to develop a data perturbation technique to generate a stress testing scenario. We apply these two proposals to two datasets to compare logistic regression and stochastic gradient boosting (SBG). We show that training a blackbox model (SGB) as conditioned on our data perturbation technique can provide insight into model performance under stressed scenarios. The empirical results show that our interpretability measure is able to capture the classification decision boundary, unlike the Gini index and the classification accuracy widely used in the banking sector. Slides Document AIAI seminar 15 June 2020 (341.1 KB / PDF) Jun 15 2020 14.00 - 15.00 15 June - Andreas Bueff Interpretable Machine Learning for Stress Testing in Credit Risk online
15 June - Andreas Bueff Link https://eu.bbcollab.com/guest/97e5b2666aa74b30870a4c83ad03760a Speaker Andreas Bueff Title Interpretable Machine Learning for Stress Testing in Credit Risk Abstract To boost the application of machine learning (ML) techniques for credit scoring models, the blackbox problem should be addressed. We propose a measure based on counterfactuals to evaluate globally the interpretability of a ML credit scoring technique. Counterfactuals assist with understanding the model with regard to classification decision boundaries. The second contribution is to develop a data perturbation technique to generate a stress testing scenario. We apply these two proposals to two datasets to compare logistic regression and stochastic gradient boosting (SBG). We show that training a blackbox model (SGB) as conditioned on our data perturbation technique can provide insight into model performance under stressed scenarios. The empirical results show that our interpretability measure is able to capture the classification decision boundary, unlike the Gini index and the classification accuracy widely used in the banking sector. Slides Document AIAI seminar 15 June 2020 (341.1 KB / PDF) Jun 15 2020 14.00 - 15.00 15 June - Andreas Bueff Interpretable Machine Learning for Stress Testing in Credit Risk online
Jun 15 2020 14.00 - 15.00 15 June - Andreas Bueff Interpretable Machine Learning for Stress Testing in Credit Risk