Fairness-Aware Race and Ethnicity Detection From Names

Xiaoxiao Shang, Zhiyuan Peng, Subramaniam Vincent, Yi Fang

Abstract: Inferring race and ethnicity from personal names can provide crucial insights for research, policy, and business. Machine learning offers a powerful tool for this inference. However, the potential biases and unfairness inherent in predictive models used for this purpose have not been explored in the literature. This paper seeks to bridge this gap by introducing Fairness-aware Race and Ethnicity Detection (FRED). We propose a novel machine learning architecture to address the unfairness by incorporating a fairness regularization term in the loss function. Furthermore, we explore the application of large language models (LLMs) for race and ethnicity prediction and tackle fairness concerns through a data pre-processing approach. We conduct comprehensive experiments on two real-world datasets with baseline comparison. The results demonstrate the effectiveness of the proposed approach.