Dataset Liver 1

Basic characteristics Liver 1

145

target objects

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

200

outlier objects

6

features

Unsupervised PCA Liver 1

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 1

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

Results Liver 1

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

590, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 58.6 ( 0.5) 58.5 ( 0.4) 58.1 ( 0.6)
Min.Cov.Determinant 58.6 ( 0.3) 58.6 ( 0.3) 59.3 ( 0.4)
Mixture of Gaussians 60.7 ( 0.6) 59.3 ( 0.7) 59.2 ( 0.2)
Naive Parzen 61.4 ( 0.7) 61.4 ( 0.7) 59.3 ( 0.5)
Parzen 59.0 ( 0.3) 58.7 ( 0.4) 58.4 ( 0.6)
k-means 57.8 ( 1.0) 56.9 ( 1.1) 56.1 ( 0.6)
1-Nearest Neighbors 59.0 ( 0.9) 60.2 ( 0.2) 58.0 ( 0.9)
k-Nearest Neighbors 59.0 ( 0.9) 60.2 ( 0.2) 58.0 ( 0.9)
knn, opt-AUC 59.0 ( 1.1) 58.7 ( 0.5) 57.3 ( 0.7)
Nearest-neighbor dist 54.6 ( 1.7) 56.7 ( 1.3) 54.6 ( 2.0)
Principal comp. 54.9 ( 0.5) 58.0 ( 1.1) 59.4 ( 2.8)
Self-Organ. Map 59.6 ( 0.7) 59.2 ( 1.3) 57.1 ( 0.9)
Auto-enc network 56.4 ( 0.9) 58.2 ( 1.0) 53.3 ( 2.5)
Spanning Tree 58.0 ( 0.9) 60.8 ( 0.5) 55.2 ( 1.7)
L_1-ball 55.4 ( 0.8) 58.2 ( 1.2) 59.7 ( 0.4)
k-center 53.7 ( 4.1) 55.9 ( 1.2) 56.5 ( 2.2)
Support vector DD 4.7 ( 1.4) 59.0 ( 0.9) 29.2 ( 3.6)
Minimax Prob. DD 58.7 ( 0.9) 59.8 ( 0.9) 58.0 ( 0.9)
LinProg DD 56.4 ( 2.6) 56.6 ( 1.2) 57.1 ( 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