Table 14 Performance on minority classes and class imbalance handling. Significant values are in bold.

From: Novel dual gland GAN architecture improves human protein localization classification using salivary and pituitary gland inspired loss functions

Model

Minority class F1 (Before)

Minority class F1 (After)

Balanced accuracy (Before)

Balanced accuracy (After)

Improvement ratio

ResNet-5047

0.762

0.881

0.812

0.904

1.156

DenseNet-12148

0.781

0.892

0.828

0.915

1.142

EfficientNet-B349

0.790

0.907

0.837

0.928

1.148

Vision Transformer50

0.778

0.913

0.825

0.932

1.173

MobileNetV351

0.741

0.858

0.794

0.885

1.158

Inception-v452

0.773

0.889

0.821

0.912

1.150

Swin Transformer53

0.796

0.918

0.841

0.936

1.153

ConvNeXt54

0.784

0.903

0.831

0.924

1.152

RegNet-Y55

0.757

0.874

0.806

0.897

1.154

NFNet56

0.803

0.925

0.846

0.941

1.152

Average

0.777

0.896

0.824

0.917

1.154

Back to article page