Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Weight of evidence transformation in credit scoring models: How does it affect the discriminatory power?

Persson, Rickard LU (2021) STAN40 20201
Department of Statistics
Abstract
Weight of evidence (WOE) transformation has been used for several decades in the credit industry. However, despite its widespread use, it has, surprisingly, been an overlooked approach in published literature. In this paper, we, therefore, investigate what effect WOE transformation has on the discriminatory power of a credit-scoring model. Our results suggest that using WOE transformation with logistic regression decreased the discriminatory power across a majority of the evaluation metrics compared to the models that did not use WOE transformed variables. Moreover, using an information value for variable selection did not provide any benefits over using the backward selection technique. However, applying support vector machine, we found... (More)
Weight of evidence (WOE) transformation has been used for several decades in the credit industry. However, despite its widespread use, it has, surprisingly, been an overlooked approach in published literature. In this paper, we, therefore, investigate what effect WOE transformation has on the discriminatory power of a credit-scoring model. Our results suggest that using WOE transformation with logistic regression decreased the discriminatory power across a majority of the evaluation metrics compared to the models that did not use WOE transformed variables. Moreover, using an information value for variable selection did not provide any benefits over using the backward selection technique. However, applying support vector machine, we found mixed results depending on the preferred evaluation metric. Using an information value seems to provide some benefits regarding variable selection compared to the recursive feature elimination technique. (Less)
Please use this url to cite or link to this publication:
author
Persson, Rickard LU
supervisor
organization
course
STAN40 20201
year
type
H1 - Master's Degree (One Year)
subject
keywords
Credit Credit-scoring WOE weight of evidence Credit-scoring models
language
English
id
9066332
date added to LUP
2021-10-20 08:54:00
date last changed
2021-10-20 08:54:00
@misc{9066332,
  abstract     = {{Weight of evidence (WOE) transformation has been used for several decades in the credit industry. However, despite its widespread use, it has, surprisingly, been an overlooked approach in published literature. In this paper, we, therefore, investigate what effect WOE transformation has on the discriminatory power of a credit-scoring model. Our results suggest that using WOE transformation with logistic regression decreased the discriminatory power across a majority of the evaluation metrics compared to the models that did not use WOE transformed variables. Moreover, using an information value for variable selection did not provide any benefits over using the backward selection technique. However, applying support vector machine, we found mixed results depending on the preferred evaluation metric. Using an information value seems to provide some benefits regarding variable selection compared to the recursive feature elimination technique.}},
  author       = {{Persson, Rickard}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Weight of evidence transformation in credit scoring models: How does it affect the discriminatory power?}},
  year         = {{2021}},
}