Dataset Balance-scale right

Basic characteristics Balance-scale right

288

target objects

The Balance Scale database from UCI. Class tip-right is used as target class. Download mat-file with Prtools dataset.

337

outlier objects

4

features

Unsupervised PCA Balance-scale right

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 Balance-scale right

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

Results Balance-scale right

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

581, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 92.9 ( 0.2) 92.9 ( 0.2) 92.9 ( 0.2)
Min.Cov.Determinant 93.0 ( 0.1) 93.0 ( 0.1) 93.0 ( 0.1)
Mixture of Gaussians 96.2 ( 0.2) 96.1 ( 0.2) 96.1 ( 0.1)
Naive Parzen 97.5 ( 0.2) 97.5 ( 0.2) 92.9 ( 0.3)
Parzen 97.3 ( 0.1) 97.1 ( 0.1) 97.3 ( 0.1)
k-means 91.3 ( 0.8) 90.9 ( 0.3) 91.4 ( 0.5)
1-Nearest Neighbors 84.4 ( 6.3) 89.4 ( 1.4) 87.5 ( 1.3)
k-Nearest Neighbors 94.5 ( 2.2) 97.0 ( 0.3) 96.4 ( 0.7)
knn, opt-AUC 94.8 ( 1.5) 96.7 ( 0.2) 96.5 ( 0.4)
Nearest-neighbor dist 84.2 ( 6.1) 88.1 ( 0.6) 85.8 ( 1.1)
Principal comp. 70.0 ( 1.9) 69.3 ( 2.4) 54.6 (10.7)
Self-Organ. Map 92.1 ( 0.4) 92.2 ( 0.9) 92.1 ( 0.4)
Auto-enc network 85.7 ( 2.2) 85.3 ( 1.9) 85.1 ( 1.3)
Spanning Tree 53.9 ( 8.2) 52.3 ( 9.6) 51.4 ( 7.6)
L_1-ball 80.6 ( 0.3) 80.6 ( 0.4) 95.5 ( 0.4)
k-center 87.9 ( 3.3) 87.4 ( 1.0) 88.3 ( 2.0)
Support vector DD 97.8 ( 0.3) 97.8 ( 0.3) 97.8 ( 0.3)
Minimax Prob. DD 97.7 ( 0.1) 97.8 ( 0.1) 97.7 ( 0.1)
LinProg DD 92.0 ( 0.3) 82.5 ( 0.3) 92.0 ( 0.3)

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