Assessing the Role of Facial Symmetry and Asymmetry between Partners in Predicting Relationship Duration: A Pilot Deep Learning Analysis of Celebrity Couples

Abstract

Prevailing studies on romantic relationships often emphasize facial symmetry as a factor in partner selection and marital satisfaction. This study aims to explore the inverse of this hypothesis—the relationship between facial dissimilarity and partnership duration among celebrity couples. Utilizing the CELEB-A dataset, which includes 202,599 images of 10,177 celebrities, we conducted an in-depth analysis using advanced artificial intelligence-based techniques. Deep learning and machine learning methods were employed to process and evaluate facial images, focusing on dissimilarity across various facial regions. Our sample comprised 1822 celebrity couples. The predictive analysis, incorporating models like Linear Regression, Ridge Regression, Random Forest, Support Vector Machine, and a Neural Network, revealed varying degrees of effectiveness in estimating partnership duration based on facial features and partnership status. However, the most notable performance was observed in Ridge Regression (Mean R2 = 0.0623 for whole face), indicating a moderate predictive capability. The study found no significant correlation between facial dissimilarity and partnership duration. These findings emphasize the complexity of predicting relationship outcomes based solely on facial attributes and suggest that other nuanced factors might play a more critical role in determining relationship dynamics. This study contributes to the understanding of the intricate nature of partnership dynamics and the limitations of facial attributes as predictors.