Dataset Abalone class 1-8

Basic characteristics Abalone class 1-8

1407

target objects

The Abalone database from UCI on the prediction of the age of abalone from physical measurements. Classes 1-8 are used as target class. Download mat-file with Prtools dataset.

2770

outlier objects

10

features

Unsupervised PCA Abalone class 1-8

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 1-8

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

Results Abalone class 1-8

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

576, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 86.1 ( 0.0) 86.2 ( 0.0) 84.0 ( 0.0)
Min.Cov.Determinant 84.8 ( 2.6) 86.0 ( 0.1) 82.2 ( 0.0)
Mixture of Gaussians 85.3 ( 0.5) 86.0 ( 0.3) 83.9 ( 0.3)
Naive Parzen 85.9 ( 0.0) 85.9 ( 0.0) 83.1 ( 0.0)
Parzen 87.3 ( 0.1) 87.7 ( 0.1) 83.1 ( 0.1)
k-means 79.2 ( 1.1) 80.1 ( 0.3) 77.9 ( 1.1)
1-Nearest Neighbors 86.5 ( 0.1) 86.2 ( 0.1) 79.2 ( 0.1)
k-Nearest Neighbors 86.5 ( 0.1) 86.2 ( 0.1) 79.2 ( 0.1)
knn, opt-AUC 87.7 ( 0.1) 86.5 ( 0.2) 79.2 ( 0.1)
Nearest-neighbor dist 58.5 ( 0.5) 58.0 ( 0.5) 51.6 ( 0.7)
Principal comp. 80.2 ( 0.1) 82.6 ( 0.1) 52.7 ( 5.5)
Self-Organ. Map 81.4 ( 0.3) 83.8 ( 0.3) 74.4 ( 0.9)
Auto-enc network 82.6 ( 0.3) 83.6 ( 0.0) 65.2 ( 4.5)
Spanning Tree 86.5 ( 0.1) NaN ( 0.0) NaN ( 0.0)
L_1-ball 77.9 ( 0.1) 75.5 ( 2.1) 82.7 ( 0.0)
k-center 76.0 ( 0.8) 76.7 ( 1.7) 70.0 ( 1.1)
Support vector DD 80.6 ( 0.1) 79.1 ( 0.2) 79.5 ( 0.1)
Minimax Prob. DD 59.4 ( 0.1) 73.5 ( 0.2) 52.0 ( 0.1)
LinProg DD 69.7 ( 0.1) 75.1 ( 0.2) 70.0 ( 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