Dataset Cancer wpbc non-ret

Basic characteristics Cancer wpbc non-ret

151

target objects

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

47

outlier objects

33

features

Unsupervised PCA Cancer wpbc non-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 non-ret

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

Results Cancer wpbc non-ret

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

554, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 59.1 (0.9) 50.8 (0.8) 52.3 (0.8)
Min.Cov.Determinant 51.1 (0.8) 51.1 (0.8) 59.0 (0.8)
Mixture of Gaussians 58.5 (1.7) 52.6 (1.6) 57.0 (0.5)
Naive Parzen 53.8 (2.2) 53.8 (2.2) 58.5 (0.8)
Parzen 58.6 (2.9) 52.2 (1.7) 58.5 (0.8)
k-means 53.6 (2.1) 52.0 (2.0) 54.0 (2.7)
1-Nearest Neighbors 59.5 (2.5) 51.7 (1.4) 58.4 (3.5)
k-Nearest Neighbors 59.5 (2.5) 51.7 (1.4) 58.4 (3.5)
Nearest-neighbor dist 55.2 (0.5) 50.1 (2.2) 55.0 (4.0)
Principal comp. 57.4 (1.8) 55.7 (1.1) 49.8 (24.2)
Self-Organ. Map 52.3 (3.0) 51.1 (2.1) 52.8 (1.0)
Auto-enc network 54.8 (3.7) 52.0 (1.1) 55.6 (7.8)
MST 61.1 (2.6) 52.7 (1.8) 52.4 (0.1)
L_1-ball 49.3 (1.8) 45.6 (1.8) 52.3 (0.8)
k-center 58.4 (5.5) 52.8 (2.0) 51.9 (5.1)
Support vector DD 49.8 (24.2) 51.7 (1.7) 57.2 (7.9)
Minimax Prob. DD 5.3 (0.1) 51.8 (1.8) 57.6 (3.0)
LinProg DD 53.9 (18.3) 53.1 (1.7) 58.6 (3.6)
Lof DD 53.9 (1.9) 49.5 (2.1) 49.1 (3.5)
Lof range DD 54.0 (1.9) 51.7 (2.1) 52.0 (2.2)
Loci DD 51.9 (2.2) 52.1 (1.6) 53.0 (1.5)

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