Table 1 Comparison of various configurations of experiments on the K-fold cross-validation (K = 10), mean and standard deviation (SD) across the folds are reported.

From: Exploiting the interplay between cross-sectional and longitudinal data in Class III malocclusion patients

 

Mean cv accuracy

SD cv accuracy

Mean cv f1 score

SD cv f1 score

Mean cv precision

SD cv precision

Mean cv recall

SD cv recall

LASSO

0.626

0.123

0.536

0.165

0.433

0.142

0.710

0.210

LASSO (TL)

0.620

0.186

0.506

0.244

0.450

0.253

0.666

0.365

LASSO (TL + FE)

0.684

0.128

0.553

0.159

0.505

0.117

0.666

0.258

GB

0.648

0.187

0.445

0.297

0.444

0.307

0.495

0.336

GB (TL)

0.782

0.147

0.628

0.287

0.627

0.305

0.650

0.300

GB (FE + TL)

0.834

0.088

0.710

0.164

0.768

0.201

0.700

0.221

  1. Gradient Boosting is more accurate than Logistic L1. The higher f1 score shows that it also balances errors in a better way. While through direct inspection of the logistic model a clear interpretation of how the prediction can be made, the accuracy and balance of the model is much poorer. The addition of TL reduces the standard deviation of the results of the GB algorithm.
Back to article page