Private fairness-aware aggregation in federated learning for financial fraud detection

Rapid adoption of artificial intelligence in financial systems has ushered in a new era of credit allocation, low-value loan risk assessment, and fraud detection, allowing banks and financial institutions to offer more personalized services. However, developing models in silos has its limitations due to unavailability of diverse data. This motivates collaboration among the financial institutions to develop a more intelligent model. However, collaboration among these institutions often introduces or amplifies existing biases, leading to unfair decision-making. In this paper, we discuss the development of a privacy-preserving personalized fairness-aware aggregation framework aimed at mitigating bias in models across different financial institutions while maintaining the inherent features they want to preserve, along with respecting their privacy. Specifically, we discuss a scenario in which the participating institutions carry biases over different attributes, while the existing state-of-the-art methods only address one attribute.