Dataset Waveform 0

Basic characteristics Waveform 0

300

target objects

The Waveform Data Generator from UCI, originally from the Classification and Regression Trees book. Class 0 is used as target class. Download mat-file with Prtools dataset.

600

outlier objects

21

features

Unsupervised PCA Waveform 0

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 Waveform 0

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

Results Waveform 0

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

598, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 84.3 ( 0.0) 84.3 ( 0.0) 81.7 ( 0.0)
Min.Cov.Determinant 84.3 ( 0.0) 84.3 ( 0.0) 81.2 ( 0.0)
Mixture of Gaussians 83.0 ( 0.6) 83.3 ( 0.5) 80.3 ( 0.4)
Naive Parzen 83.8 ( 0.0) 83.8 ( 0.0) 80.8 ( 0.0)
Parzen 82.5 ( 0.0) 83.0 ( 0.0) 79.4 ( 0.0)
k-means 82.5 ( 0.6) 82.0 ( 0.9) 80.1 ( 1.1)
1-Nearest Neighbors 81.3 ( 0.0) 82.0 ( 0.0) 78.0 ( 0.0)
k-Nearest Neighbors 81.3 ( 0.0) 82.0 ( 0.0) 78.0 ( 0.0)
knn, opt-AUC 81.9 ( 0.8) 83.9 ( 0.4) 78.8 ( 1.3)
Nearest-neighbor dist 76.4 ( 0.0) 76.6 ( 0.0) 73.4 ( 0.0)
Principal comp. 74.2 ( 0.0) 74.5 ( 0.0) 58.6 ( 0.0)
Self-Organ. Map 82.4 ( 0.3) 82.4 ( 0.5) 78.8 ( 0.4)
Auto-enc network 81.2 ( 0.1) 76.8 ( 0.0) 78.4 ( 0.1)
Spanning Tree 81.7 ( 0.0) NaN ( 0.0) 77.9 ( 0.0)
L_1-ball 81.8 ( 0.0) 80.6 ( 0.0) 76.9 ( 0.0)
k-center 62.5 (27.7) 80.8 ( 1.4) 76.0 ( 2.7)
Support vector DD 81.4 ( 0.0) 82.7 ( 0.0) 78.2 ( 0.0)
Minimax Prob. DD 81.5 ( 0.0) 82.6 ( 0.0) 78.2 ( 0.0)
LinProg DD 86.7 ( 0.0) 85.9 ( 0.0) 84.3 ( 0.0)

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