Skip to main content
. 2018 Apr 16;6:e4568. doi: 10.7717/peerj.4568

Table 5. Consolidated results of Kruskal–Wallis H and post-hoc tests.

Metric Kruskal-Wallis H summary Mean ranks Post-hoc
Accuracy χ2(5) = 15.508, p = 0.008 AlexNet 11.20 Xception & ResNet-50 (p = 0.005)
Xception & VGG-16 (p = 0.007)
Customized & ResNet-50 (p = 0.017)
VGG-16 22.30
ResNet-50 23.00
Xception 7.20
DenseNet-121 19.60
Customized 9.70
AUC χ2(5) = 18.958, p = 0.002 AlexNet 13.00 Xception & ResNet-50 (p = 0.034)
Xception & VGG-16 (p = 0.030)
Xception & DenseNet-121 (p = 0.014)
VGG-16 21.70
ResNet-50 21.50
Xception 4.50
DenseNet-121 22.90
Customized 9.40
Sensitivity χ2(5) = 5.518, p = 0.356 AlexNet 16.20
VGG-16 17.30
ResNet-50 15.80
Xception 11.40
DenseNet-121 21.80
Customized 10.50
Specificity χ2(5) = 6.639, p = 0.249 AlexNet 9.80
VGG-16 20.70
ResNet-50 21.30
Xception 13.30
DenseNet-121 14.10
Customized 13.80
F1-score χ2(5) = 14.798, p = 0.011 AlexNet 11.70 Xception & ResNet-50 (p = 0.005)
Xception & VGG-16 (p = 0.006)
Xception & DenseNet-121 (p = 0.023)
VGG-16 22.20
ResNet-50 22.60
Xception 6.90
DenseNet-121 19.50
Customized 10.10
MCC χ2(5) = 14.487, p = 0.013 AlexNet 11.30 Xception & ResNet-50 (p = 0.007)
Xception & VGG-16 (p = 0.008)
Xception & DenseNet-121 (p = 0.034)
Customized & ResNet-50 (p = 0.021)
VGG-16 22.30
ResNet-50 22.60
Xception 7.60
DenseNet-121 19.40
Customized 9.80

Notes.

Bold text indicate the performance measures of the best-performing model/s.