Dataset Wine 1

Basic characteristics Wine 1

59

target objects

The Wine Recognition database from UCI, to determine the origin of wines using chemical analysis. Class 1 is used as target class. Download mat-file with Prtools dataset.

119

outlier objects

13

features

Unsupervised PCA Wine 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 Wine 1

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

Results Wine 1

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

595, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 97.1 ( 0.3) 98.8 ( 0.4) 95.7 ( 0.5)
Min.Cov.Determinant 96.3 ( 0.9) 96.3 ( 0.9) 95.6 ( 0.6)
Mixture of Gaussians 96.6 ( 0.6) 98.8 ( 0.1) 95.7 ( 0.6)
Naive Parzen 99.7 ( 0.3) 99.7 ( 0.3) 95.0 ( 0.7)
Parzen 91.0 ( 1.4) 99.5 ( 0.2) 95.0 ( 0.7)
k-means 92.5 ( 1.6) 99.6 ( 0.1) 92.2 ( 1.4)
1-Nearest Neighbors 90.4 ( 1.2) 99.5 ( 0.2) 87.8 ( 1.4)
k-Nearest Neighbors 90.4 ( 1.2) 99.5 ( 0.2) 87.8 ( 1.4)
knn, opt-AUC 90.4 ( 1.2) 99.4 ( 0.1) 93.1 ( 1.3)
Nearest-neighbor dist 75.5 ( 1.7) 99.3 ( 0.6) 70.0 ( 4.7)
Principal comp. 81.3 ( 1.7) 91.5 ( 2.5) 53.5 (34.3)
Self-Organ. Map 96.2 ( 0.4) 99.4 ( 0.2) 95.0 ( 0.3)
Auto-enc network 87.9 ( 1.8) 97.2 ( 0.7) 94.0 ( 0.7)
Spanning Tree 93.2 ( 0.6) NaN ( 0.0) 57.8 (15.9)
L_1-ball 89.7 ( 3.3) 96.9 ( 0.3) 95.7 ( 0.5)
k-center 89.6 ( 2.3) 99.0 ( 0.3) 83.8 ( 3.8)
Support vector DD 6.9 ( 0.8) 51.2 (28.6) 31.3 ( 7.4)
Minimax Prob. DD 74.9 ( 4.8) 99.5 ( 0.2) 78.3 ( 4.2)
LinProg DD 41.5 (17.0) 99.7 ( 0.1) 42.2 (13.8)

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