Abstract
Many studies have been conducted on the components of facial attractiveness and sexual dimorphism. However, most previous studies were mainly laboratory experiments based on experimenters' hypotheses, making it difficult to comprehensively and in detail examine the relationship between various facial features, sexual dimorphism, and facial attractiveness. Therefore, in this study, I employed a data-driven approach that does not rely on hypotheses. Specifically, using the various facial features and facial impression scores in the facial image dataset, I extracted features important for facial attractiveness and sexual dimorphism using a random forest regression model. Then I investigated causal relationships using a statistical causal discovery method called LinGAM. The results showed that for male images, various facial features predicted attractiveness via sexual dimorphism, whereas, for female images, various facial features predicted sexual dimorphism via attractiveness. These suggested that, from a data-driven perspective, the relationship between sexual dimorphism and attractiveness varies by gender.
