Calibration of Machine Learning Classifiers for Probability of Default Modelling
This work addresses the need for better calibration in credit scoring for financial institutions, though it is incremental as it builds on existing calibration methods.
The study tackled the problem of improving probability calibration in credit scoring models by comparing calibration techniques like Platt Scaling and Isotonic Regression on real-world datasets, finding that re-calibration with Isotonic Regression in time-series data enhances long-term calibration and helps non-parametric models outperform Logistic Regression in Brier Score Loss.
Binary classification is highly used in credit scoring in the estimation of probability of default. The validation of such predictive models is based both on rank ability, and also on calibration (i.e. how accurately the probabilities output by the model map to the observed probabilities). In this study we cover the current best practices regarding calibration for binary classification, and explore how different approaches yield different results on real world credit scoring data. The limitations of evaluating credit scoring models using only rank ability metrics are explored. A benchmark is run on 18 real world datasets, and results compared. The calibration techniques used are Platt Scaling and Isotonic Regression. Also, different machine learning models are used: Logistic Regression, Random Forest Classifiers, and Gradient Boosting Classifiers. Results show that when the dataset is treated as a time series, the use of re-calibration with Isotonic Regression is able to improve the long term calibration better than the alternative methods. Using re-calibration, the non-parametric models are able to outperform the Logistic Regression on Brier Score Loss.