Table 4.
Paired-sample t test of the area under the receiver operating characteristic curve.
| Test result pair(s) | Asymptotic | AUCa difference | SE differenceb | Asymptotic 95% CI | ||
| z | Sig. (2-tailed P value)c,d | Lower bound | Upper bound | |||
| XGBTe-RFf | –2.74 | .006 | –0.043 | 0.179 | –0.073 | –0.012 |
| XGBT-LRg | 6.118 | <.001 | 0.146 | 0.209 | 0.099 | 0.193 |
| XGBT-C5h | –2.967 | .003 | –0.054 | 0.175 | –0.09 | –0.018 |
| RF-LR | 7.405 | <.001 | 0.189 | 0.197 | 0.139 | 0.239 |
| RF-C5 | –0.655 | .51 | –0.011 | 0.161 | –0.046 | 0.023 |
| LR-C5 | –8.094 | <.001 | –0.201 | 0.193 | –0.249 | –0.152 |
aAUC: area under the curve.
bUnder the nonparametric assumption.
cNull hypothesis: true area difference = 0.
dA P value <0.008 is statistically significant (using Bonferroni correction). P values were derived from the pair-wise corrected resampled t test.
eXGBT: extreme gradient boosting tree.
fRF: random forest.
gLR: logistic regression.
hC5: C5.0 decision tree.