Dataset Glass headlamps

Basic characteristics Glass headlamps

29

target objects

The Glass Identification database from UCI, originally to distinguish 6 types of glass by their oxide content. The headlamps class is used as target class. Download mat-file with Prtools dataset.

185

outlier objects

9

features

Unsupervised PCA Glass headlamps

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 Glass headlamps

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

Results Glass headlamps

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

587, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 86.1 ( 1.9) 84.3 ( 2.1) 84.7 ( 1.4)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 82.9 ( 1.7)
Mixture of Gaussians 79.7 ( 2.2) 83.7 ( 1.0) 80.1 ( 2.4)
Naive Parzen 87.2 ( 2.0) 87.2 ( 2.0) 86.3 ( 1.7)
Parzen 80.0 ( 1.8) 85.5 ( 0.8) 80.7 ( 1.2)
k-means 83.6 ( 3.1) 88.7 ( 1.8) 83.3 ( 3.3)
1-Nearest Neighbors 78.8 ( 2.0) 85.0 ( 1.4) 78.2 ( 2.4)
k-Nearest Neighbors 78.8 ( 2.0) 85.0 ( 1.4) 78.2 ( 2.4)
knn, opt-AUC 79.3 ( 2.2) 84.8 ( 0.9) 82.6 ( 3.3)
Nearest-neighbor dist 28.3 ( 3.0) 65.6 ( 1.6) 26.1 ( 2.7)
Principal comp. 84.0 ( 2.7) 86.1 ( 2.3) 83.5 ( 4.1)
Self-Organ. Map 79.6 ( 2.1) 85.0 ( 1.4) 78.9 ( 2.7)
Auto-enc network 74.7 ( 4.2) 87.4 ( 2.5) 83.3 ( 1.5)
Spanning Tree 78.6 ( 2.3) 84.2 ( 1.2) 77.4 ( 2.6)
L_1-ball 53.7 ( 0.6) 84.4 ( 1.5) 84.6 ( 3.9)
k-center 77.9 ( 2.4) 84.2 ( 2.0) 77.6 ( 2.5)
Support vector DD 75.3 ( 5.7) 52.3 ( 4.4) 72.7 ( 4.4)
Minimax Prob. DD 78.9 ( 2.1) 85.0 ( 1.3) 78.4 ( 2.5)
LinProg DD 81.8 ( 1.8) 85.4 ( 0.6) 80.6 ( 1.0)

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