Dataset Vowel 10

Basic characteristics Vowel 10

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 10 is used as target class. Download mat-file with Prtools dataset.

480

outlier objects

10

features

Unsupervised PCA Vowel 10

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 10

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

Results Vowel 10

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

611, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 96.8 ( 0.0) 96.7 ( 0.0) 71.4 ( 0.0)
Min.Cov.Determinant 95.5 ( 0.0) 95.5 ( 0.0) 54.6 ( 0.0)
Mixture of Gaussians 91.7 ( 0.5) 92.2 ( 1.6) 70.9 ( 7.3)
Naive Parzen 64.7 ( 0.0) 64.7 ( 0.0) 76.9 ( 0.0)
Parzen 91.1 ( 0.0) 91.7 ( 0.0) 69.8 ( 0.0)
k-means 88.6 ( 1.7) 88.8 ( 1.9) 67.3 ( 2.2)
1-Nearest Neighbors 91.1 ( 0.0) 91.7 ( 0.0) 69.9 ( 0.0)
k-Nearest Neighbors 91.1 ( 0.0) 91.7 ( 0.0) 69.9 ( 0.0)
knn, opt-AUC 91.1 ( 0.0) 91.7 ( 0.0) 69.9 ( 0.0)
Nearest-neighbor dist 69.2 ( 0.0) 74.7 ( 0.0) 70.2 ( 0.0)
Principal comp. 95.7 ( 0.0) 94.0 ( 0.0) 72.0 ( 0.0)
Self-Organ. Map 94.1 ( 1.6) 92.9 ( 1.2) 73.8 ( 1.0)
Auto-enc network 95.5 ( 0.5) 94.0 ( 0.0) 68.0 ( 6.6)
Spanning Tree 92.5 ( 0.0) 92.8 ( 0.0) 75.9 ( 0.0)
L_1-ball 74.7 ( 0.0) 82.8 ( 0.0) 65.2 ( 0.0)
k-center 90.7 ( 0.3) 91.7 ( 0.3) 67.9 ( 1.9)
Support vector DD 91.9 ( 0.2) 53.6 ( 5.7) 70.9 ( 1.0)
Minimax Prob. DD 92.2 ( 0.0) 91.7 ( 0.0) 73.0 ( 0.0)
LinProg DD 91.3 ( 0.0) 92.5 ( 0.0) 74.0 ( 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