Dataset Diabetes (targetcl. present)

Basic characteristics Diabetes (targetcl. present)

500

target objects

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

268

outlier objects

8

features

Unsupervised PCA Diabetes (targetcl. present)

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. present)

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

Results Diabetes (targetcl. present)

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

517, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 70.5 (0.3) 72.1 (0.3) 71.1 (0.2)
Min.Cov.Determinant 71.5 (0.2) 71.5 (0.2) 70.2 (0.2)
Mixture of Gaussians 67.4 (0.3) 73.8 (0.3) 67.6 (0.3)
Naive Parzen 67.9 (0.3) 67.8 (0.3) 67.9 (0.5)
Parzen 67.6 (0.4) 75.6 (0.2) 66.1 (0.4)
k-means 65.9 (0.7) 71.2 (1.0) 64.7 (0.9)
1-Nearest Neighbors 66.7 (0.7) 72.1 (0.2) 60.6 (1.1)
k-Nearest Neighbors 66.7 (0.7) 72.1 (0.2) 60.6 (1.1)
Nearest-neighbor dist 55.1 (1.0) 54.4 (0.9) 51.1 (0.8)
Principal comp. 58.7 (0.2) 64.0 (0.6) 67.4 (0.7)
Self-Organ. Map 69.2 (0.7) 70.9 (0.9) 68.9 (0.9)
Auto-enc network 59.8 (1.8) 65.8 (0.5) 65.7 (0.9)
MST 65.9 (0.7) 71.5 (0.3) 59.9 (0.8)
L_1-ball 60.8 (1.1) 71.1 (0.5) 66.8 (0.2)
k-center 60.6 (1.6) 67.8 (0.9) 58.5 (1.7)
Support vector DD 57.7 (9.8) 73.2 (0.5) 59.0 (2.6)
Minimax Prob. DD 65.6 (0.7) 72.9 (0.3) 60.5 (1.2)
LinProg DD 66.8 (0.7) 63.4 (0.5) 61.1 (1.0)
Lof DD 56.0 (0.9) 63.6 (1.1) 57.2 (1.2)
Lof range DD 63.1 (0.6) 67.6 (0.8) 63.6 (0.7)
Loci DD 66.1 (0.5) 69.2 (0.5) 66.4 (0.4)

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