Open Access

Huyen Giang Thi Thu, Thang Viet Doan, Ha-Bang Ban, Tai Le Quy

• The digitalization of credit scoring has become essential for financial institutions and commercial banks, especially in the era of digital transformation. Machine learning (ML) techniques are commonly used to evaluate customers’ creditworthiness. However, the predicted outcomes of ML models can be biased toward protected attributes, such as race or gender. Numerous fairness-aware ML models and fairness measures have been proposed in recent years. However, their behavior in the context of credit scoring has not been thoroughly investigated. In this paper, we present a comprehensive experimental study of fairness-aware ML for credit scoring. Our study examines several key aspects of the problem, including financial datasets, predictive models, and fairness measures. In addition, we analyze structural dependencies between protectedThe digitalization of credit scoring has become essential for financial institutions and commercial banks, especially in the era of digital transformation. Machine learning (ML) techniques are commonly used to evaluate customers’ creditworthiness. However, the predicted outcomes of ML models can be biased toward protected attributes, such as race or gender. Numerous fairness-aware ML models and fairness measures have been proposed in recent years. However, their behavior in the context of credit scoring has not been thoroughly investigated. In this paper, we present a comprehensive experimental study of fairness-aware ML for credit scoring. Our study examines several key aspects of the problem, including financial datasets, predictive models, and fairness measures. In addition, we analyze structural dependencies between protected attributes and the class label using a Bayesian network to better understand statistical relationships within the datasets. We further provide a detailed evaluation of fairness-aware predictive models and fairness measures on widely used credit scoring datasets. The experimental results show that fairness-aware models achieve a better balance between predictive accuracy and fairness than traditional classification models.…