Fig. 3: RMSE behaviour across GAIN training iterations.

Throughout the training iterative process, the imputation accuracy of GAIN is monitored using RMSE for both training (blue line) and test (red line) subsets, where a proportion of 10% of their data is intermittently obscured to simulate missing data. The trends illustrate the GAIN stabilisation, achieving final RMSE values of 0.11 ± 0.05 for the training set and 0.19 ± 0.02 for the test set.