Skip to main content
. 2020 Aug 5;20:55. doi: 10.1186/s40644-020-00329-8

Fig. 5.

Fig. 5

Performance drop comparision across all class boundaries, shown as mean and 95% confidene intervals, calculated with the adjusted bootstrap percentile (BCa) method. Lower is better. The drop for the models using robust features is lower for all recorded metrics, except for the sensitivity. Values are reported in Table 3