Dataset Breast Wisconsin (targetcl. malignant)

Basic characteristics Breast Wisconsin (targetcl. malignant)

241

target objects

The original database of the Wisconsin Breast Cancer Databases from UCI, containing 699 instances, collected between 1989 and 1991. Missing values have been replaced by -1.000000. Download mat-file with Prtools dataset.

458

outlier objects

9

features

Unsupervised PCA Breast Wisconsin (targetcl. malignant)

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 Breast Wisconsin (targetcl. malignant)

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

Results Breast Wisconsin (targetcl. malignant)

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

504, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 82.3 (0.2) 82.2 (0.2) 86.6 (0.3)
Min.Cov.Determinant 73.5 (0.1) 73.5 (0.1) 78.3 (0.3)
Mixture of Gaussians 69.1 (3.2) 66.0 (3.0) 75.1 (3.1)
Naive Parzen 96.5 (0.4) 96.5 (0.4) 90.7 (0.8)
Parzen 72.3 (0.5) 68.1 (0.5) 77.1 (0.6)
k-means 84.6 (3.5) 86.9 (3.2) 86.0 (4.2)
1-Nearest Neighbors 69.3 (0.6) 63.0 (0.9) 74.5 (0.7)
k-Nearest Neighbors 69.3 (0.6) 63.0 (0.9) 74.5 (0.7)
Nearest-neighbor dist 63.6 (2.2) 56.9 (1.7) 65.4 (2.4)
Principal comp. 30.3 (1.0) 34.6 (0.9) 40.3 (0.9)
Self-Organ. Map 79.0 (2.3) 71.4 (1.3) 82.6 (1.4)
Auto-enc network 38.4 (0.9) 29.1 (1.1) 43.2 (0.8)
MST 70.6 (1.7) 64.7 (0.8) 74.0 (0.8)
L_1-ball 96.1 (0.3) 97.4 (0.2) 58.4 (4.2)
k-center 71.5 (12.3) 72.6 (13.6) 70.9 (9.5)
Support vector DD 66.1 (0.8) 70.3 (0.5) 74.3 (0.7)
Minimax Prob. DD 69.4 (0.6) 69.8 (0.7) 74.5 (0.7)
LinProg DD 80.0 (0.5) 94.8 (0.4) 83.2 (0.6)
Lof DD 81.1 (2.0) 84.1 (2.0) 83.0 (1.7)
Lof range DD 80.7 (0.9) 83.6 (0.7) 82.1 (1.1)
Loci DD 98.1 (0.1) 97.9 (0.1) 98.2 (0.0)

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