Nor is color blindness exclusively a male phenomena; in Northern Europeans, 8% of males are colorblind while 0.5% of females are. Suppose a few more sexual dim

Nor is color blindness exclusively a male phenomena; in Northern Europeans, 8% of males are colorblind while 0.5% of females are. Suppose a few more sexual dimorphic traits like this exist in the eyes; perhaps differences that have no practical effect on human vision and have consequently gone unnoticed by clinicians. If the ML model is picking up a few of these dimorphic traits, it could perhaps classify sex with more accuracy than anybody looking at a single trait could. This is pretty standard Bayesian stuff; it's the way basic "Plan for Spam" style Bayesian spam filters work.