Practical bias correction in neural networks: A credit default prediction case study
| Sieuwert van Otterloo |
Artificial Intelligence
Snel, P. and van Otterloo, S. 2022. Practical bias correction in neural networks: A credit default prediction case study, Computers and Society Research Journal (2022), 3
https://doi.org/10.54822/BEWO3288
Piet Snel and Sieuwert van Otterloo
Email: sieuwert@ictinstitute.nl
Article published online 11 May 2022.
Abstract
Artificial intelligence (AI) is increasingly being used for decision-making. Technological developments have significantly increased the performance of Artificial intelligence (AI) models but have also increased their complexity. As a result, IT professionals are struggling to develop fair AI implementations as (1) measuring fairness in a practical case is difficult due to multiple definitions (2) literature on this topic is complex especially when multiple types of bias occur and (3) lack of practical cases in which corrections are made. Using a case study, we demonstrate how both gender and age bias can be addressed in practice. We do this by developing a credit default prediction model and detecting and mitigating both age and gender bias within this model. A neural network was trained using a real world credit data set. Existing ‘bias’ in the data set and bias introduced by this initial model was measured using a combination of methods. A corrected model was created by training and evaluating a series of models, to control bias along multiple dimensions. The final model eliminates the measured bias without sacrificing accuracy. It uses a top-down post-processing technique focusing on an equal increase of the default rate per group. Our paper concludes with recommendations for how unfair bias can be avoided in real world applications.
Key words: Machine Learning, Bias, Credit Default Prediction, Fairness, Taiwanese Credit Data, Fair AI
PDF Download
This paper has been published in the “Computer and Society Research Journal“, a free to publish, open access journal for socially relevant computer science research.
We have also created a summary presentation for general audiences as part of our summer school at the Utrecht University of Applied Sciences.
Code & Data Set
The code used to construct our models is available to download as Python Jupyter notebooks. Please cite this paper if you use it.
We use the data set published in: I. C. Yeh and C. H. Lien. The comparisons of data mining techniques for the predictive accuracy of probability of default of credit card clients. Expert Systems with Applications, 36(2):2473–2480, 2009, https://doi.org/10.1016/j.eswa.2007.12.020. The data set is available to download via the University of California, Irvine.
How to cite this paper
We recommend you to use the following code to cite the paper:
@article{Snel2022,
title={Practical bias correction in neural networks: a credit default prediction case study},
author={Snel, P. and van Otterloo, S.},
journal={Computers and Society Research Journal},
number={3},
year={2022}}
}
It should look similar to this, depending on your template:
Snel and van Otterloo – Practical bias correction in neural networks: a credit default prediction case study, Computers and Society Research Journal (2022), 1
Dr. Sieuwert van Otterloo is a court-certified IT expert with interests in agile, security, software research and IT-contracts.