Dataset Cancer wpbc ret

Basic characteristics Cancer wpbc ret

47

target objects

The new prognostic database of the Wisconsin Breast Cancer Databases from @UCIurl@, containing 198 instances. Original feature 3 (recurrence Time or disease-free time) is included in the features. Class returning is used as target class. Download mat-file with Prtools dataset.

151

outlier objects

33

features

Unsupervised PCA Cancer wpbc ret

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 Cancer wpbc ret

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

Results Cancer wpbc ret

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

555, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 61.7 (0.9) 63.0 (1.4) 53.8 (1.6)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 49.4 (1.0)
Mixture of Gaussians 52.9 (1.3) 59.1 (1.6) 51.2 (1.1)
Naive Parzen 58.9 (1.8) 59.0 (1.9) 49.2 (1.3)
Parzen 52.8 (2.7) 59.4 (1.9) 49.2 (1.3)
k-means 51.0 (1.2) 59.4 (3.0) 47.8 (2.8)
1-Nearest Neighbors 53.1 (2.8) 59.2 (2.2) 45.2 (2.1)
k-Nearest Neighbors 53.1 (2.8) 59.2 (2.2) 45.2 (2.1)
Nearest-neighbor dist 59.1 (1.5) 47.6 (2.7) 56.1 (2.9)
Principal comp. 50.3 (1.6) 63.3 (2.0) 50.2 (24.2)
Self-Organ. Map 53.5 (1.5) 60.8 (3.5) 53.3 (1.5)
Auto-enc network 48.0 (5.0) 61.8 (2.8) 42.5 (3.5)
MST 56.5 (2.8) 59.2 (1.4) 51.0 (0.1)
L_1-ball 51.8 (1.6) 59.9 (2.4) 53.8 (1.6)
k-center 51.2 (2.6) 59.5 (2.3) 42.1 (4.1)
Support vector DD 50.2 (24.2) 59.6 (1.4) 51.3 (15.7)
Minimax Prob. DD 0.0 (0.0) 59.1 (2.0) 39.5 (2.0)
LinProg DD 49.2 (23.6) 60.4 (1.7) 46.0 (1.2)
Lof DD 54.3 (2.2) 56.4 (3.6) 51.0 (3.5)
Lof range DD 49.1 (3.2) 56.6 (2.3) 46.4 (2.4)
Loci DD 53.1 (1.1) 60.0 (2.4) 52.1 (1.1)

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