Table 2 The deep ROC analysis of the machine learning models in stratified 5-repeated 5-fold cross-validation.
From: Machine learning-based diagnostic prediction of minimal change disease: model development study
FPR | [0,1] | [0.0.33] | [0.33,0.67] | [0.67,1] |
|---|---|---|---|---|
Predicted probability | All | High | Medium | Low |
TabPFN | ||||
AUROCni | 0.915 (0.047) | 0.894 (0.050) | 0.923 (0.083) | 0.995 (0.024) |
Avg sensitivity | 0.915 (0.047) | 0.766 (0.115) | 0.977 (0.032) | 1 (0.001) |
Avg specificity | 0.915 (0.047) | 0.942 (0.025) | 0.344 (0.291) | 0.012 (0.059) |
LightGBM | ||||
AUROCni | 0.911 (0.041) | 0.887 (0.052) | 0.941 (0.069) | 0.979 (0.048) |
Avg sensitivity | 0.911 (0.041) | 0.757 (0.108) | 0.976 (0.031) | 0.998 (0.006) |
Avg specificity | 0.911 (0.041) | 0.933 (0.033) | 0.244 (0.287) | 0.041 (0.097) |
Random forest | ||||
AUROCni | 0.906 (0.043) | 0.882 (0.051) | 0.922 (0.075) | 0.985 (0.041) |
Avg sensitivity | 0.906 (0.043) | 0.742 (0.110) | 0.974 (0.029) | 0.999 (0.004) |
Avg specificity | 0.906 (0.043) | 0.933 (0.030) | 0.296 (0.294) | 0.0324 (0.090) |
Artificial neural network | ||||
AUROCni | 0.880 (0.057) | 0.864 (0.055) | 0.866 (0.082) | 0.964 (0.073) |
Avg sensitivity | 0.880 (0.057) | 0.698 (0.117) | 0.945 (0.055) | 0.996 (0.010) |
Avg specificity | 0.880 (0.057) | 0.929 (0.026) | 0.470 (0.197) | 0.062 (0.114) |
Logistic regression | ||||
AUROCni | 0.888 (0.059) | 0.883 (0.047) | 0.868 (0.086) | 0.946 (0.102) |
Avg sensitivity | 0.888 (0.059) | 0.734 (0.107) | 0.937 (0.065) | 0.992 (0.018) |
Avg specificity | 0.888 (0.059) | 0.941 (0.022) | 0.397 (0.239) | 0.058 (0.110) |
