Dataset Abalone class 11-29

Basic characteristics Abalone class 11-29

1447

target objects

The Abalone database from UCI on the prediction of the age of abalone from physical measurements. Classes 11-29 are used as target class. Download mat-file with Prtools dataset.

2730

outlier objects

10

features

Unsupervised PCA Abalone class 11-29

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 Abalone class 11-29

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

Results Abalone class 11-29

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

578, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 65.9 ( 0.0) 63.5 ( 0.1) 68.0 ( 0.0)
Min.Cov.Determinant 62.5 ( 0.1) 62.2 ( 0.6) 71.5 ( 0.0)
Mixture of Gaussians 64.2 ( 1.9) 55.1 ( 1.1) 66.3 ( 1.2)
Naive Parzen 66.4 ( 0.1) 66.4 ( 0.1) 66.2 ( 0.1)
Parzen 46.7 ( 0.2) 47.8 ( 0.1) 64.0 ( 0.2)
k-means 61.5 ( 1.2) 63.5 ( 1.3) 61.6 ( 1.2)
1-Nearest Neighbors 44.8 ( 0.2) 45.9 ( 0.1) 61.1 ( 0.3)
k-Nearest Neighbors 44.8 ( 0.2) 45.9 ( 0.1) 61.1 ( 0.3)
knn, opt-AUC 44.8 ( 0.2) 46.1 ( 0.3) 61.1 ( 0.3)
Nearest-neighbor dist 55.4 ( 0.3) 53.7 ( 0.4) 52.5 ( 0.5)
Principal comp. 56.7 ( 0.0) 54.0 ( 0.0) 52.7 ( 8.3)
Self-Organ. Map 61.8 ( 0.7) 57.1 ( 0.5) 63.0 ( 0.7)
Auto-enc network 55.8 ( 1.2) 57.4 ( 0.1) 61.0 ( 1.3)
Spanning Tree 47.9 ( 1.4) 45.9 ( 0.1) 50.8 ( 0.2)
L_1-ball 68.7 ( 0.2) 51.8 ( 2.2) 48.3 ( 0.5)
k-center 55.6 ( 0.8) 54.0 ( 1.2) 55.9 ( 1.0)
Support vector DD 64.5 ( 0.1) 62.6 ( 0.4) 57.1 ( 0.1)
Minimax Prob. DD 63.3 ( 0.2) 58.1 ( 0.4) 53.1 ( 0.1)
LinProg DD 63.6 ( 0.2) 69.4 ( 0.2) 55.0 ( 0.1)

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