Dataset Vehicle van

Basic characteristics Vehicle van

199

target objects

The vehicle dataset from Statlog, to recognize a vehicle from its silhouette. Class van is used as target class. Download mat-file with Prtools dataset.

647

outlier objects

18

features

Unsupervised PCA Vehicle van

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 Vehicle van

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

Results Vehicle van

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

612, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 95.6 ( 0.1) 96.3 ( 0.1) 78.3 ( 0.2)
Min.Cov.Determinant 95.0 ( 0.1) 95.0 ( 0.1) 80.2 ( 0.4)
Mixture of Gaussians 95.0 ( 0.1) 96.1 ( 0.3) 84.6 ( 0.4)
Naive Parzen 86.9 ( 0.2) 86.9 ( 0.2) 83.3 ( 0.5)
Parzen 47.1 ( 8.6) 85.3 ( 4.9) 54.8 ( 4.6)
k-means 88.9 ( 0.6) 91.6 ( 0.2) 85.6 ( 0.5)
1-Nearest Neighbors 70.6 ( 8.9) 95.2 ( 0.3) 89.7 ( 0.2)
k-Nearest Neighbors 70.6 ( 8.9) 95.2 ( 0.3) 89.7 ( 0.2)
knn, opt-AUC 73.3 ( 6.4) 95.2 ( 0.3) 89.8 ( 0.2)
Nearest-neighbor dist 91.6 ( 2.2) 93.1 ( 0.3) 86.7 ( 0.5)
Principal comp. 90.3 ( 0.8) 92.8 ( 0.4) 78.1 ( 0.3)
Self-Organ. Map 84.6 ( 0.8) 93.2 ( 0.2) 79.7 ( 0.1)
Auto-enc network 90.5 ( 0.7) 90.8 ( 0.2) 82.7 ( 0.6)
Spanning Tree 95.1 ( 2.6) 95.7 ( 0.3) 89.9 ( 0.2)
L_1-ball 65.4 ( 1.0) 78.2 ( 0.7) 75.9 ( 0.9)
k-center 72.5 ( 6.6) 91.0 ( 0.6) 85.8 ( 0.8)
Support vector DD 0.3 ( 0.0) 45.9 ( 6.9) 8.5 ( 1.0)
Minimax Prob. DD 94.3 ( 0.1) 95.7 ( 0.3) 88.6 ( 0.3)
LinProg DD 38.6 ( 9.7) 92.4 ( 0.2) 41.9 ( 5.0)

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