Fig. 4.

Sex classification AUROC across 1 - 300 independent CNNs included in the ensemble model. The raw data is plotted, as well as the projection of this trend using a logistics growth model ($y=\frac{a}{1+b{e}^{-kx}},k>0)$, which assumes a hard upper limit ($a$) to the classification accuracy that can be achieved by simply increasing the number of models in the ensemble. The model predicts that simply adding more models to the ensemble beyond 300 achieves limited returns. The upper limit is 0.8477, with 95% confidence bounds between 0.8473 and 0.8481.