Dataset Diabetes (targetcl. absent)

Basic characteristics Diabetes (targetcl. absent)

268

target objects

The Pima Indians Diabetes Database from UCI. Download mat-file with Prtools dataset.

500

outlier objects

8

features

Unsupervised PCA Diabetes (targetcl. absent)

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 Diabetes (targetcl. absent)

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

Results Diabetes (targetcl. absent)

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

518, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 56.6 (0.2) 53.1 (0.3) 57.8 (0.2)
Min.Cov.Determinant 52.1 (0.2) 52.1 (0.2) 55.7 (0.4)
Mixture of Gaussians 56.2 (0.4) 52.9 (0.7) 53.3 (0.1)
Naive Parzen 56.6 (0.4) 56.6 (0.4) 57.5 (0.3)
Parzen 54.5 (0.5) 47.0 (0.4) 56.4 (0.1)
k-means 57.2 (0.9) 53.1 (1.1) 59.2 (0.6)
1-Nearest Neighbors 51.8 (0.5) 48.5 (0.3) 55.2 (0.8)
k-Nearest Neighbors 51.8 (0.5) 48.5 (0.3) 55.2 (0.8)
Nearest-neighbor dist 49.1 (0.9) 48.3 (2.0) 55.8 (1.3)
Principal comp. 59.6 (1.0) 46.6 (0.6) 63.8 (0.3)
Self-Organ. Map 53.7 (1.8) 50.2 (1.7) 58.8 (1.2)
Auto-enc network 55.7 (1.8) 50.7 (0.8) 56.8 (1.3)
MST 51.6 (0.6) 48.1 (0.4) 53.8 (0.8)
L_1-ball 50.7 (0.2) 59.2 (0.5) 56.4 (0.2)
k-center 53.5 (1.3) 51.3 (2.3) 51.5 (0.9)
Support vector DD 43.8 (9.6) 49.6 (0.8) 54.0 (1.6)
Minimax Prob. DD 50.6 (0.5) 49.5 (0.4) 55.3 (0.7)
LinProg DD 52.0 (0.4) 59.7 (0.4) 55.2 (0.7)
Lof DD 58.5 (1.1) 57.6 (0.9) 48.8 (1.5)
Lof range DD 59.8 (0.7) 53.0 (0.7) 61.8 (1.3)
Loci DD 58.1 (0.3) 51.4 (0.5) 58.7 (0.8)

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