Skip to main content
. 2020 Feb 27;27(4):621–633. doi: 10.1093/jamia/ocz228

Figure 3.

Figure 3.

Reliability diagrams of test set estimates produced by logistic regression (LR) and support vector machine (SVM) models grouped for the Hosmer-Lemeshow (H-L) C-statistics and the H-L H-statistics. Data points of estimates produced by the models and their actual binary outcomes are plotted to show the distribution of the actual data. The Alive outcome is indicated as 0 and the Deceased outcome is indicated as 1. Corresponding H-L statistics and P values are shown in the graphs. (A) LR and SVM estimates grouped for the calculation of the H-L C-statistic. The number of patients within each bin is the same (n = 125). (B) Estimates grouped for the calculation of the H-L H-statistic. The number of patients in each group is shown in the graph. Both graphs show that the SVM underestimates the actual number of deaths, as shown by the red line deviating from the diagonal line. LR is relatively well calibrated (blue line).

HHS Vulnerability Disclosure