Figure 4.

Results of the multiple comparison tests for the classification accuracy of NeurDNet across different scenarios. The term “Prob” in the vertical axis refers to the probabilistic features extracted from the first-stage classifier. The vertical axis represents different testing scenarios and the horizontal one represents the classification accuracy. Also note that the circles denote the mean classification accuracy and the lines define the range of the $95\%$ confidence interval. Please note that in all of the plots, the performance of Prob QDA (in blue) is compared with other scenarios. Any overlap between the lines of two scenarios corroborates that the performance of NeurDNet is not significantly altered by changing one hyper-parameter to another. The plots include significance tests for [portion of test set—patients’ visits accounted]: (a) $25\%$—all visits; (b) $35\%$—all visits; (c) $25\%$—first visits; (d) $35\%$—first visits.