Dataset Liver 2

Basic characteristics Liver 2

200

target objects

The Liver-disorders database from UCI. Test for liver disorders that might arise from excessive alcohol consumption. Disorder absence is used as target class. Download mat-file with Prtools dataset.

145

outlier objects

6

features

Unsupervised PCA Liver 2

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 Liver 2

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

Results Liver 2

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

591, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 50.7 ( 0.6) 50.9 ( 0.5) 48.7 ( 0.6)
Min.Cov.Determinant 51.8 ( 0.4) 51.8 ( 0.4) 50.0 ( 0.5)
Mixture of Gaussians 50.5 ( 0.7) 49.4 ( 0.6) 48.4 ( 0.4)
Naive Parzen 48.4 ( 0.8) 48.4 ( 0.8) 48.5 ( 0.7)
Parzen 49.6 ( 0.7) 46.9 ( 0.8) 48.1 ( 0.9)
k-means 50.0 ( 0.4) 46.9 ( 1.4) 49.5 ( 0.5)
1-Nearest Neighbors 50.1 ( 1.1) 51.1 ( 0.7) 49.9 ( 0.6)
k-Nearest Neighbors 50.1 ( 1.1) 51.1 ( 0.7) 49.9 ( 0.6)
knn, opt-AUC 50.1 ( 1.1) 50.1 ( 0.9) 49.5 ( 1.3)
Nearest-neighbor dist 54.4 ( 1.3) 56.6 ( 1.0) 48.9 ( 1.5)
Principal comp. 55.9 ( 0.9) 60.8 ( 0.8) 53.0 (11.9)
Self-Organ. Map 54.3 ( 0.7) 48.7 ( 1.7) 53.8 ( 0.9)
Auto-enc network 52.5 ( 3.6) 60.8 ( 0.8) 48.7 ( 2.8)
Spanning Tree 49.9 ( 1.0) 51.4 ( 0.5) 49.9 ( 1.3)
L_1-ball 54.4 ( 0.8) 47.6 ( 0.5) 46.6 ( 0.5)
k-center 51.5 ( 2.8) 48.3 ( 0.6) 48.4 ( 2.3)
Support vector DD 0.7 ( 0.4) 49.6 ( 1.0) 31.8 ( 3.7)
Minimax Prob. DD 50.1 ( 1.1) 52.1 ( 1.1) 49.8 ( 0.5)
LinProg DD 50.3 ( 0.6) 50.6 ( 0.5) 49.2 ( 1.6)

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