Table 3 Classification with leave-one-out cross validation on the first wave dataset. The bold font identifies the models with best performance, for each of the different rows (i.e. for each predictive model type).

From: Early outcome detection for COVID-19 patients

 

F1 score

Accuracy

  

Coverage 90%

Coverage 75%

 

Coverage 90%

Coverage 75%

Model

GA

DT-RFE

LR-RFE

DT-RFE

LR-RFE

GA

DT-RFE

LR-RFE

DT-RFE

LR-RFE

LR

0.899

0.824

0.865

0.809

0.806

0.903

0.829

0.870

0.820

0.810

DT

0.863

0.766

0.808

0.830

0.734

0.870

0.785

0.819

0.832

0.731

RF

0.874

0.818

0.808

0.846

0.776

0.876

0.825

0.815

0.851

0.778

NB

0.778

0.800

0.836

0.775

0.803

0.762

0.801

0.839

0.789

0.806

SVM

0.840

0.640

0.771

0.824

0.718

0.859

0.748

0.799

0.845

0.736

Support

185

246

254

161

216

185

246

254

161

216

  1. Each row corresponds to a different model type. We compare models trained on variables selected by our feature selection method (GA) with those selected by the two recursive feature elimination algorithms with the two different coverage thresholds (DT-RFE, LR-RFE).
Back to article page