Measuring Sexual Preference With Mouse Tracking and Machine Learning

Various methods have been proposed to measure a person’s sexual preference, but most of these methods have not been evaluated on a yet unseen sample of participants. We examine here the importance of using an unseen sample with the recently introduced mouse tracking method, which has been previously shown to produce reliable group differences in trajectories. To this end, we applied machine learning on data from a mouse tracking task with 52 self‐declared heterosexual male and female participants, who either responded according to their sexual preference or were asked to ‘fake’ their response and always provide the opposite response from their preference. While performance was high when the model was applied to the same data it was trained on (AUC = 0.91), substantially lower classification performance was found on unseen data (AUC = 0.69). Our results show that statistically reliable group differences do not automatically lead to reliable predictions in unseen data. Before a method can be used in clinical practice, it should therefore first be tested on unseen data.