A comparison of the performance between two of our approaches for combining ground truths using metrics from Sections 3.3.1 and 3.3.3. We evaluate the performance of redshift predictions averaged over the redshift range 0.3 < z < 1.5. The two chosen approaches are transfer learning (NN-TL) and composite dataset training (NN-Combo). NN-Combo performs better than NN-TL across both datasets on LSST-like metrics.