Dataset Hepatitis (targetcl. live)

Basic characteristics Hepatitis (targetcl. live)

123

target objects

The hepatitis database from UCI to predict if the patient will live or die.Missing values have been replaced by the mean of the feature. Download mat-file with Prtools dataset.

32

outlier objects

19

features

Unsupervised PCA Hepatitis (targetcl. live)

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 Hepatitis (targetcl. live)

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

Results Hepatitis (targetcl. live)

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

516, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 63.6 (1.7) 82.1 (1.0) 49.8 (1.9)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 54.7 (1.9)
Mixture of Gaussians 58.9 (1.7) 78.3 (1.0) 55.8 (1.8)
Naive Parzen 80.1 (0.7) 80.1 (0.7) 55.8 (2.5)
Parzen 60.1 (2.3) 79.0 (1.0) 58.2 (1.4)
k-means 51.1 (3.4) 79.7 (1.4) 49.8 (2.6)
1-Nearest Neighbors 59.4 (2.5) 78.6 (1.0) 50.6 (2.0)
k-Nearest Neighbors 59.4 (2.5) 78.6 (1.0) 50.6 (2.0)
Nearest-neighbor dist 50.5 (3.6) 51.6 (2.7) 47.9 (4.6)
Principal comp. 56.4 (0.9) 79.1 (1.6) 5.6 (1.8)
Self-Organ. Map 51.7 (2.2) 77.6 (1.1) 48.1 (3.2)
Auto-enc network 55.2 (7.6) 80.0 (1.2) 51.2 (6.9)
MST 57.0 (1.6) 79.1 (1.6) 49.7 (0.9)
L_1-ball 69.5 (1.7) 69.6 (2.5) 48.1 (3.0)
k-center 56.4 (2.7) 78.6 (4.7) 54.6 (3.6)
Support vector DD 4.3 (0.7) 78.7 (1.1) 32.4 (0.6)
Minimax Prob. DD 56.1 (2.1) 78.7 (0.9) 50.1 (1.8)
LinProg DD 58.4 (2.3) 80.6 (1.4) 50.8 (1.7)
Lof DD 55.3 (3.2) 46.2 (2.2) 41.8 (1.9)
Lof range DD 57.3 (2.2) 66.8 (0.6) 52.3 (1.6)
Loci DD 56.7 (1.5) 81.7 (1.2) 54.5 (2.1)

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