Table 2 REFINED-CNN performance comparison with competing models for the GDSC dataset.

From: Representation of features as images with neighborhood dependencies for compatibility with convolutional neural networks

Model

NRMSE (20%)

NRMSE (50%)

NRMSE (80%)

PCC (20%)

PCC (50%)

PCC (80%)

Bias (20%)

Bias (50%)

Bias (80%)

EN

0.890

0.889

0.887

0.488

0.484

0.486

0.848

0.849

0.840

RF

0.609

0.620

0.569

0.797

0.785

0.821

0.433

0.417

0.337

SVR

0.750

0.742

0.525

0.847

0.845

0.853

0.257

0.273

0.241

ANN

1.407

0.475

0.435

0.519

0.883

0.901

0.784

0.153

0.233

Random-CNN

0.579

0.456

0.441

0.836

0.892

0.903

0.215

0.193

0.222

PCA-CNN

0.612

0.461

0.443

0.820

0.891

0.901

0.201

0.228

0.179

REFINED-CNN

0.541

0.439

0.414

0.845

0.899

0.911

0.255

0.173

0.197

  1. The numbers in parentheses indicate the percentages of the available data used for training.
  2. EN elastic net, RF random forest, SVR support vector regression, ANN artificial neural networks, Random-CNN random mapping based convolutional neural network, PCA-CNN principal component analysis based convolutional neural networks, REFINED-CNN proposed REFINED approach-based convolutional neural networks. Bold values indicate the best performances.
Back to article page