Dataset Balance-scale middle

Basic characteristics Balance-scale middle

49

target objects

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

576

outlier objects

4

features

Unsupervised PCA Balance-scale middle

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 middle

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

Results Balance-scale middle

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

580, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 93.1 ( 0.6) 93.1 ( 0.6) 36.1 ( 1.1)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 35.7 ( 1.3)
Mixture of Gaussians 83.0 ( 1.0) 83.1 ( 0.9) 36.5 ( 2.2)
Naive Parzen 20.9 ( 3.6) 20.9 ( 3.6) 40.9 ( 3.0)
Parzen 69.6 ( 0.8) 69.8 ( 0.8) 30.5 ( 1.4)
k-means 58.8 ( 3.4) 58.3 ( 3.1) 35.4 ( 2.0)
1-Nearest Neighbors 53.9 (10.1) 61.2 ( 1.2) 12.7 ( 1.1)
k-Nearest Neighbors 75.2 ( 6.2) 75.6 ( 1.4) 49.9 ( 1.2)
knn, opt-AUC 69.3 ( 8.5) 73.1 ( 2.0) 44.7 ( 3.2)
Nearest-neighbor dist 49.8 ( 9.2) 56.0 ( 1.1) 12.0 ( 1.0)
Principal comp. 94.1 ( 0.7) 94.4 ( 0.8) 46.8 (19.8)
Self-Organ. Map 60.6 ( 2.4) 62.5 ( 1.3) 25.6 ( 4.6)
Auto-enc network 91.5 ( 1.2) 90.6 ( 1.6) 42.7 ( 3.6)
Spanning Tree 48.7 ( 3.4) 48.5 ( 6.4) 50.8 ( 3.1)
L_1-ball 49.2 ( 6.1) 55.4 ( 4.5) 42.4 ( 2.2)
k-center 62.8 ( 5.9) 65.8 ( 3.9) 33.6 ( 2.8)
Support vector DD 68.2 ( 0.6) 74.1 ( 1.8) 24.2 ( 1.6)
Minimax Prob. DD 68.4 ( 1.2) 77.8 ( 1.0) 19.7 ( 0.7)
LinProg DD 71.2 ( 0.5) 63.5 ( 0.9) 38.4 ( 0.6)

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