Dataset Vowel 3

Basic characteristics Vowel 3

48

target objects

The Vowel undocumented dataset from UCI. Speaker independent recognition of the eleven steady state vowels of British English using a specified training set of lpc derived log area ratios. Vowel 3 is used as target class. Download mat-file with Prtools dataset.

480

outlier objects

10

features

Unsupervised PCA Vowel 3

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 Vowel 3

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

Results Vowel 3

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

604, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 94.3 ( 0.0) 94.1 ( 0.0) 94.5 ( 0.0)
Min.Cov.Determinant 90.7 ( 0.0) 90.7 ( 0.0) 95.3 ( 0.0)
Mixture of Gaussians 96.3 ( 0.6) 95.3 ( 0.9) 93.4 ( 1.9)
Naive Parzen 91.8 ( 0.0) 91.8 ( 0.0) 89.6 ( 0.0)
Parzen 96.7 ( 0.0) 94.3 ( 0.0) 93.3 ( 0.0)
k-means 94.6 ( 1.4) 94.0 ( 0.8) 90.0 ( 0.9)
1-Nearest Neighbors 96.7 ( 0.0) 94.2 ( 0.0) 93.3 ( 0.0)
k-Nearest Neighbors 96.7 ( 0.0) 94.2 ( 0.0) 93.3 ( 0.0)
knn, opt-AUC 96.7 ( 0.2) 94.2 ( 0.0) 93.3 ( 0.0)
Nearest-neighbor dist 87.8 ( 0.0) 80.8 ( 0.0) 82.3 ( 0.0)
Principal comp. 95.7 ( 0.0) 91.6 ( 0.0) 91.3 ( 0.0)
Self-Organ. Map 96.6 ( 0.5) 95.7 ( 0.5) 93.6 ( 0.9)
Auto-enc network 95.8 ( 0.0) 91.6 ( 0.0) 83.3 ( 0.5)
Spanning Tree 97.1 ( 0.0) 94.3 ( 0.0) 94.2 ( 0.0)
L_1-ball 83.4 ( 0.0) 89.1 ( 0.0) 91.5 ( 0.0)
k-center 95.6 ( 0.6) 94.3 ( 0.3) 92.7 ( 1.5)
Support vector DD 97.3 ( 0.3) 44.1 ( 5.8) 95.2 ( 0.3)
Minimax Prob. DD 97.0 ( 0.0) 94.6 ( 0.0) 94.5 ( 0.0)
LinProg DD 94.6 ( 0.0) 95.1 ( 0.0) 92.6 ( 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