Table 3 Mean areas under the ROC curve from a ten-fold-cross-validation using either a linear model (perceptron) or multilayer perceptrons (MLPs) with 1 and 10 hidden units, respectively, on all features.

From: Machine learning-based tool to assess risk of hemodynamically significant PDA in extremely premature infants

 

AUC

 

Mean

± Standard deviation

P (t-test)

Linear perceptron

0.662

0.058

 

MLP with 3 HU

0.604

0.077

0.0189

MLP with 10 HU

0.541

0.104

<0.001

Forward selection

0.638

 

0.080

Backward elimination

0.645

0.077

0.270

  1. P values refer to a t-test comparing the performance with that of the linear perceptron.
Back to article page