A1310
Title: Effects of Imbalanced Datasets on Algorithmic Fairness in Credit Scoring
Authors: Yujia Chen - University of Edinburgh Business School (United Kingdom) [presenting]
Raffaella Calabrese - University of Edinburgh (United Kingdom)
Abstract: The growing use of AI in credit markets leverages powerful algorithms to process vast data, hence improving the classification between good-type and bad-type borrowers. However, AI development has raised concerns about discrimination biases, particularly in credit scoring, where models may disadvantage groups based on protected attributes such as gender, age, or race. Class imbalance is another common problem in credit scoring, as the number of observations in the minority class (defaults) is much less than that in the majority class (non defaults). Current works have focused on the (adverse) effect of imbalanced datasets on the predictive ability and interpretation performance of machine learning techniques. However, no research examines the effects of imbalanced datasets on the fairness of machine learning algorithms. We propose an experimental framework with novel fairness metrics to measure the algorithmic fairness of credit scoring models on different levels of class imbalance.
