Dataset Glass vehicle float

Basic characteristics Glass vehicle float

17

target objects

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

197

outlier objects

9

features

Unsupervised PCA Glass vehicle float

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 vehicle float

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

Results Glass vehicle float

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

584, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 73.4 ( 2.2) 88.0 ( 2.3) 67.5 ( 0.4)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 69.5 ( 1.1)
Mixture of Gaussians 74.7 ( 2.8) 78.1 ( 1.4) 70.8 ( 2.8)
Naive Parzen 72.7 ( 2.2) 72.7 ( 2.2) 71.1 ( 1.5)
Parzen 75.7 ( 2.0) 76.4 ( 1.3) 70.0 ( 1.8)
k-means 76.1 ( 4.9) 75.3 ( 3.3) 71.7 ( 8.1)
1-Nearest Neighbors 77.1 ( 2.3) 77.8 ( 2.3) 71.8 ( 2.9)
k-Nearest Neighbors 77.1 ( 2.3) 77.8 ( 2.3) 71.8 ( 2.9)
knn, opt-AUC 75.7 ( 2.1) 76.1 ( 2.3) 68.5 ( 2.8)
Nearest-neighbor dist 77.5 ( 2.3) 82.8 ( 1.6) 68.2 ( 2.7)
Principal comp. 75.0 ( 2.3) 79.8 ( 1.3) 68.9 ( 5.7)
Self-Organ. Map 77.1 ( 2.2) 78.3 ( 2.5) 72.1 ( 3.1)
Auto-enc network 73.0 ( 3.6) 80.9 ( 4.9) 70.5 ( 3.8)
Spanning Tree 77.5 ( 2.1) 79.5 ( 2.2) 73.4 ( 2.5)
L_1-ball 52.0 ( 4.5) 67.0 ( 3.5) 64.4 ( 2.0)
k-center 77.1 ( 2.3) 77.8 ( 2.1) 71.4 ( 2.4)
Support vector DD 70.4 ( 7.7) 77.9 ( 3.0) 65.5 ( 7.2)
Minimax Prob. DD 79.4 ( 1.2) 77.9 ( 2.2) 74.0 ( 1.2)
LinProg DD 74.7 ( 2.2) 77.9 ( 2.1) 70.4 ( 1.8)

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