Skip to main content

Table 4 Comparison of accuracy of projection methods

From: ANINet: a deep neural network for skull ancestry estimation

Internet AlexNet (%) Vgg-16 (%) GoogLenet (%) Resnet-50 (%) DenseNet-121 (%) SqueezeNet (%) ANINet (%)
Local projection        
Accuracy 90.83 62.17 95.92 96.33 97.97 92.91 98.04
Precision 89.38 69.70 96.11 96.74 97.98 93.76 98.76
Recall 93.17 57.76 96.58 98.09 98.15 93.58 98.44
F1-score 91.97 63.23 96.65 97.15 98.19 93.70 98.27
Specificity 87.46 67.89 95.01 95.09 98.55 90.88 99.01
Global projection        
Accuracy 95.04 60.78 96.08 98.03 98.09 95.91 98.21
Precision 92.26 64.94 97.31 98.06 98.15 96.89 98.49
Recall 96.27 57.68 94.41 98.38 98.43 96.47 98.51
F1-score 94.80 61.17 95.96 98.50 98.70 96.80 98.77
Specificity 92.01 62.31 93.69 97.98 98.85 93.00 99.09
Combining two projection methods        
Accuracy 95.56 65.00 98.22 98.30 98.36 96.28 99.03
Precision 92.97 69.92 97.32 98.49 98.57 97.04 98.82
Recall 96.83 58.26 96.97 98.51 98.77 96.87 98.86
F1-score 94.93 64.21 98.03 98.74 98.85 97.84 98.93
Specificity 94.15 68.88 96.33 98.79 98.99 94.69 99.11
  1. The significance of bold means that in a certain aspect (performance, model size), this model is the best among these models