A Replication Study: Machine Learning Models Are Capable of Predicting Sexual Orientation From Facial Images

A Replication Study: Machine Learning Models Are Capable of Predicting Sexual Orientation From Facial Images. John Leuner. Master's thesis. Feb 2017, https://arxiv.org/pdf/1902.10739.pdf

Abstract: Recent research used machine learning methods to predict a persons sexual orientationfrom their photograph (Wang and Kosinski, 2017). To verify this result, two of thesemodels are replicated, one based on a deep neural network (DNN) and one on facialmorphology (FM). Using a new dataset of 20,910 photographs from dating websites, theability to predict sexual orientation is confirmed (DNN accuracy male 68%, female 77%,FM male 62%, female 72%). To investigate whether facial features such as brightness orpredominant colours are predictive of sexual orientation, a new model trained on highlyblurred facial images was created. This model was also able to predict sexual orienta-tion (male 63%, female 72%). The tested models are invariant to intentional changesto a subjects makeup, eyewear, facial hair and head pose (angle that the photograph istaken at). It is shown that the head pose is not correlated with sexual orientation. Whiledemonstrating that dating profile images carry rich information about sexual orientationthese results leave open the question of how much is determined by facial morphologyand how much by differences in grooming, presentation and lifestyle. The advent ofnew technology that is able to detect sexual orientation in this way may have seriousimplications for the privacy and safety of gay men and women.