Skip to main content

Table 21 WHO-5 classification results

From: Predicting subjective well-being in a high-risk sample of Russian mental health app users

Classification

Threshold

N (Classes)

Features

Best model

F1-macro

F1-weighted

F1-low

F1-high

TruePositiveRate (low)

FalsePositiveRate (low)

binary

0.51

221/151

Words

AdaBoost

0.56

0.581

0.669

0.452

0.697

0.57

RuLIWC

DecisionTree

0.571

0.582

0.631

0.512

0.611

0.457

AppCats

AdaBoost

0.58

0.602

0.694

0.466

0.738

0.57

Behavior

DecisionTree

0.543

0.559

0.63

0.456

0.638

0.55

Clusters

RandomForest

0.539

0.571

0.714

0.363

0.832

0.715

binary majority baseline

  

0.378

0.456

0.373

0

1

1

trinary

0.35/0.59

111/158/103

Words

AdaBoost

0.44

0.447

0.407

0.43

0.378

0.195

RuLIWC

AdaBoost

0.381

0.399

0.413

0.238

0.405

0.241

AppCats

AdaBoost

0.422

0.443

0.402

0.294

0.396

0.241

Behavior

AdaBoost

0.425

0.438

0.427

0.329

0.414

0.23

Clusters

DecisionTree

0.358

0.364

0.338

0.339

0.351

0.295

clusters + RuLIWC + Words

AdaBoost

0.483

0.493

0.502

0.433

0.45

0.161

trinary majority baseline

  

0.199

0.253

–

–

0

0