Dataset Biomed (targetcl. normal)

Basic characteristics Biomed (targetcl. normal)

127

target objects

The purpose of the analysis is to develop a screening procedure to detect carriers and to describe its effectiveness. Entries with missing values have been removed. Download mat-file with Prtools dataset.

67

outlier objects

5

features

Unsupervised PCA Biomed (targetcl. normal)

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 Biomed (targetcl. normal)

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

Results Biomed (targetcl. normal)

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

511, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 90.0 (0.4) 89.9 (0.5) 88.1 (0.8)
Min.Cov.Determinant 89.8 (0.8) 89.8 (0.8) 88.0 (1.0)
Mixture of Gaussians 91.2 (0.9) 91.1 (0.8) 89.0 (1.0)
Naive Parzen 93.1 (0.2) 93.1 (0.2) 88.7 (1.3)
Parzen 90.0 (1.1) 91.5 (0.9) 88.6 (1.0)
k-means 87.8 (1.2) 90.2 (0.9) 85.4 (1.5)
1-Nearest Neighbors 89.1 (0.8) 91.4 (1.2) 88.3 (1.0)
k-Nearest Neighbors 89.1 (0.8) 91.4 (1.2) 88.3 (1.0)
Nearest-neighbor dist 72.7 (1.9) 83.2 (0.4) 69.3 (2.8)
Principal comp. 89.7 (0.5) 77.6 (3.1) 83.1 (0.7)
Self-Organ. Map 88.7 (0.8) 90.8 (0.6) 87.7 (0.8)
Auto-enc network 85.6 (2.2) 89.0 (1.3) 86.4 (1.1)
MST 89.3 (1.1) 91.4 (1.2) 88.4 (0.5)
L_1-ball 80.9 (0.6) 86.9 (0.4) 83.1 (0.8)
k-center 87.8 (2.4) 90.6 (1.5) 86.6 (0.8)
Support vector DD 52.2 (0.3) 91.5 (0.9) 65.1 (0.8)
Minimax Prob. DD 85.8 (0.8) 90.9 (1.0) 85.7 (0.9)
LinProg DD 87.5 (0.9) 88.9 (0.8) 86.8 (1.0)
Lof DD 89.1 (0.8) 89.3 (1.0) 84.0 (0.8)
Lof range DD 88.1 (1.0) 90.3 (0.8) 85.4 (1.0)
Loci DD 88.1 (1.1) 90.6 (0.9) 87.6 (0.7)

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