Skip to main content

Table 2 Adjusted Rand Index, Normalised Mutual Information, and Fowlkes-Mallows Index computed for our proposed greedy-graph and brute-graph feature selection methods, as well as classification-based, phylogeny-based, and leave-one-variable-out methods

From: Feature graphs for interpretable unsupervised tree ensembles: centrality, interaction, and application in disease subtyping

 

Greedy-graph

Brute-graph

Classification-based

Phylogeny-based

Leave-one–variable-out

 

mean

mean

p-value

mean

p-value

 

mean

p-value

 

mean

p-value

 

Adjusted Rand Index

   Iris

0.8202

0.8201

1.000

0.8051

1.000

 

0.8074

1.000

 

0.8756

0.092

 

   Liver disorders

0.0328

0.0342

1.000

0.0307

1.000

 

0.0323

1.000

 

0.0108

<0.001

\(\uparrow\)

   Ecoli

0.3565

0.3540

1.000

0.3209

<0.001

\(\uparrow\)

0.3420

0.867

 

0.3687

1.000

 

   Breast tissue

0.3455

0.3444

1.000

0.3273

<0.001

\(\uparrow\)

0.3455

1.000

 

0.3524

1.000

 

   Glass

0.2183

0.2163

1.000

0.2047

0.012

\(\uparrow\)

0.1921

<0.001

\(\uparrow\)

0.1483

<0.001

\(\uparrow\)

   Wine

0.5778

0.5687

0.131

0.6534

<0.001

\(\downarrow\)

0.4279

<0.001

\(\uparrow\)

0.5055

<0.001

\(\uparrow\)

   Lymphography

0.1197

0.1186

1.000

0.0888

<0.001

\(\uparrow\)

0.0770

<0.001

\(\uparrow\)

0.0950

<0.001

\(\uparrow\)

   Parkinson

0.2087

0.2035

0.459

0.1028

<0.001

\(\uparrow\)

0.0911

<0.001

\(\uparrow\)

0.2905

<0.001

\(\downarrow\)

   Ionosphere

0.1253

-

-

0.0800

<0.001

\(\uparrow\)

0.1125

1.000

 

0.1913

<0.001

\(\downarrow\)

   Sonar

0.0217

-

-

0.0030

<0.001

\(\uparrow\)

0.0198

0.210

 

0.0432

<0.001

\(\downarrow\)

   Monotonicity

0.6806

  

0.6459

<0.001

\(\uparrow\)

0.6051

<0.001

\(\uparrow\)

0.6500

0.070

 

Normalised Mutual Information

   Iris

0.8067

0.8072

1.000

0.7905

1.000

 

0.7923

1.000

 

0.8549

0.092

 

   Liver disorders

0.0295

0.0306

1.000

0.0270

1.000

 

0.0288

1.000

 

0.0142

<0.001

\(\uparrow\)

   Ecoli

0.4388

0.4381

1.000

0.4337

0.069

 

0.4440

1.000

 

0.3845

<0.001

\(\uparrow\)

   Breast tissue

0.5076

0.5092

1.000

0.4984

<0.001

\(\uparrow\)

0.5258

<0.001

\(\downarrow\)

0.5191

<0.001

\(\downarrow\)

   Glass

0.3065

0.3086

1.000

0.2942

<0.001

\(\uparrow\)

0.2915

<0.001

\(\uparrow\)

0.2141

<0.001

\(\uparrow\)

   Wine

0.5854

0.5824

1.000

0.6668

<0.001

 

0.4399

<0.001

\(\uparrow\)

0.5245

<0.001

\(\uparrow\)

   Lymphography

0.1157

0.1167

1.000

0.0938

<0.001

\(\uparrow\)

0.0974

<0.001

\(\uparrow\)

0.1099

<0.001

\(\uparrow\)

   Parkinson

0.1155

0.1154

1.000

0.1325

<0.001

 

0.1217

1.000

 

0.1992

<0.001

\(\downarrow\)

   Ionosphere

0.1150

-

-

0.0727

<0.001

\(\uparrow\)

0.1223

0.065

 

0.1942

<0.001

\(\downarrow\)

   Sonar

0.0577

-

-

0.0073

<0.001

\(\uparrow\)

0.0169

<0.001

\(\uparrow\)

0.0494

<0.001

\(\uparrow\)

   Monotonicity

0.6975

  

0.6393

<0.001

\(\uparrow\)

0.6229

<0.001

\(\uparrow\)

0.6930

1.000

 

Fowlkes-Mallows Index

   Iris

0.8759

0.8799

1.0000

0.8712

1.000

 

0.8701

1.000

 

0.9062

1.000

 

   Liver disorders

0.5349

0.5352

1.0000

0.5302

1.000

 

0.5344

1.000

 

0.5488

<0.001

\(\downarrow\)

   Ecoli

0.5023

0.4998

1.0000

0.4797

<0.001

\(\uparrow\)

0.5009

1.000

 

0.5758

<0.001

\(\downarrow\)

   Breast tissue

0.4611

0.4595

1.0000

0.4516

0.044

\(\uparrow\)

0.4614

1.000

 

0.4605

1.000

 

   Glass

0.4558

0.4597

0.1105

0.4452

<0.001

\(\uparrow\)

0.4347

<0.001

\(\uparrow\)

0.4123

<0.001

\(\uparrow\)

   Wine

0.7165

0.7168

1.0000

0.7758

<0.001

\(\downarrow\)

0.6307

<0.001

\(\uparrow\)

0.6798

<0.001

\(\uparrow\)

   Lymphography

0.4603

0.4575

1.0000

0.4292

<0.001

\(\uparrow\)

0.4094

<0.001

\(\uparrow\)

0.4360

<0.001

\(\uparrow\)

   Parkinson

0.7215

0.7186

1.0000

0.6183

<0.001

\(\uparrow\)

0.6177

<0.001

\(\uparrow\)

0.7410

<0.001

\(\downarrow\)

   Ionosphere

0.6717

-

-

0.6752

0.145

 

0.7014

<0.001

\(\downarrow\)

0.6774

0.824

 

   Sonar

0.5987

-

-

0.5298

<0.001

\(\uparrow\)

0.5285

<0.001

\(\uparrow\)

0.5588

<0.001

\(\uparrow\)

   Monotonicity

0.6291

  

0.5855

0.005

\(\uparrow\)

0.5722

<0.001

\(\uparrow\)

0.5955

0.004

\(\uparrow\)

  1. We also report p-values for Wilcoxon signed-rank tests with Bonferroni correction, evaluating our greedy-graph method against each other methods. Symbols \(\uparrow\) (\(\downarrow\)) indicate whether the greedy-graph method outperforms (underperforms) the other method with statistical significance. Missing values indicate cases where feature selection could not be applied due to excessive computation times (> 48 hours)