Dataset Metas 2

Basic characteristics Metas 2

99

target objects

The Metas145 gene expression dataset, presented in Van 't Veer et al., Gene expression profiling predicts clinical outcome of breast cancer, Nature, 2002. Download mat-file with Prtools dataset.

46

outlier objects

4919

features

Unsupervised PCA Metas 2

On the left, the PCA scatterplot is shown, on the right the retained variance for varying number of features.
On the left, the PCA scatterplot is shown of data rescaled to unit variance, on the right the retained variance.

Supervised Fisher Metas 2

On the left, the Fisher scatterplot is shown, on the right the ROC curve along this direction.

Results Metas 2

The experiments are performed using dd_tools. A rudimentary explanation of the classifiers is given in the classifier section.

575, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 69.1 ( 1.5) 68.1 ( 1.1) 59.9 ( 2.3)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) NaN ( 0.0)
Mixture of Gaussians NaN ( 0.0) NaN ( 0.0) NaN ( 0.0)
Naive Parzen 65.3 ( 0.8) 65.4 ( 0.9) 63.0 ( 1.7)
Parzen 64.8 (21.5) 64.8 (21.5) 57.1 ( 0.7)
k-means 65.9 ( 1.2) 65.9 ( 1.6) 57.5 ( 0.5)
1-Nearest Neighbors 65.1 ( 2.2) 65.0 ( 1.2) 56.5 ( 1.3)
k-Nearest Neighbors 65.1 ( 2.2) 65.0 ( 1.2) 56.5 ( 1.3)
knn, opt-AUC 64.9 ( 2.0) 65.3 ( 1.0) 59.1 ( 2.6)
Nearest-neighbor dist 63.0 ( 1.3) 60.8 ( 2.1) 58.6 ( 1.4)
Principal comp. 68.9 ( 1.4) 67.6 ( 1.0) 60.9 ( 5.3)
Self-Organ. Map 65.2 ( 1.7) 65.7 ( 0.5) 55.9 ( 1.3)
Auto-enc network NaN ( 0.0) NaN ( 0.0) 58.1 ( 3.0)
Spanning Tree NaN ( 0.0) 65.0 ( 1.0) 53.6 ( 1.6)
L_1-ball 61.2 ( 1.7) 51.3 ( 1.9) 53.4 ( 4.5)
k-center 62.6 ( 1.5) 63.2 ( 1.1) 53.0 ( 2.9)
Support vector DD 64.8 (21.5) 64.8 (21.5) 64.8 (21.5)
Minimax Prob. DD 0.0 ( 0.0) 0.0 ( 0.0) 0.0 ( 0.0)
LinProg DD 64.8 (21.5) 64.8 (21.5) 64.8 (21.5)

Classifier projection spaces The first classifier projection spaces are obtained by computing the classifier label disagreements (setting the threshold on 10% target error) and applying an MDS on the resulting distance matrix between classifiers:



Original



Unit variance



PCA mapped

Classifier projection spaces The second versions of the classifier projection spaces are obtained by computing the classifier ranking disagreements and applying an MDS on the resulting distance matrix between classifiers:



Original



Unit variance



PCA mapped