Racially Inclusive Approach to Facial Beauty Modeling Using Machine Learning
Erik Nguyen, Sampson Akwafuo, Doina Bein, Blessing Ojeme · 2024
Facial beauty perception is a complex area of study that has intrigued researchers across various disciplines. While some argue that it is subjective, influenced by personal and cultural factors, others propose that it is objective and rooted in evolutionary biology. This study explores the latter perspective, aiming to model facial beauty with an emphasis on racial fairness. Departing from black box convolutional deep learning approaches that are susceptible to racial biases, particularly arising from their holistic consideration of facial attributes such as skin tone, our focus lies solely on designing a more transparent machine learning model that integrates guardrails to prevent the introduction of such biases. By deliberately excluding skin tone and selecting specific features for the model to learn from, we aim to ensure a more equitable assessment across diverse racial and ethnic groups. Following rigorous training and evaluation, our hybrid model demonstrated impressive predictive performance, despite prioritizing transparency and racial fairness over complexity.