Combating Algorithmic Bias: responsible fairness measures, algorithms and toolkits in retail banking

Fairness has gained unprecedented support in a world of daily emerging scientific inquisition and
discovery, aiming to tackle algorithmic bias effectively. Extensive efforts have been devoted to defining
and embodying what is bias (discrimination) and developing tools that enable machine learning
practitioners to detect and mitigate bias during algorithm design. However, mysteries are yet to be
solved on the practical application of these fairness measures and toolkits. This research proposal
presents a systematic review of identified algorithmic bias issues and the proposed fairness solution
space, focusing on the development of novel approaches to attain fairness in the banking system. The
general objective is broken down into three sub-objectives. The first involves creating fairness
assessments on real banking datasets and implementing existing advanced toolkits to measure bias.
The second sub-objective focuses on deploying appropriate measures of fairness specific to the
available datasets, including group, individual, and causality-based fairness, while the third subobjective
aims to design novel approaches customized to stakeholders and the banking system to
achieve fairness. The methodologies outlined in this proposal offer a comprehensive approach to
measure and mitigate biases in the banking system. By addressing these issues, the insights gained are
to foster collaboration between practitioners and fairness experts, ultimately facilitating the
development of practical and user-friendly fair ML toolkits.