Dataset Abalone class 9-10

Basic characteristics Abalone class 9-10

1323

target objects

The Abalone database from UCI on the prediction of the age of abalone from physical measurements. Classes 9-10 are used as target class. Download mat-file with Prtools dataset.

2854

outlier objects

10

features

Unsupervised PCA Abalone class 9-10

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 Abalone class 9-10

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

Results Abalone class 9-10

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

577, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 66.8 ( 0.0) 65.1 ( 0.0) 63.5 ( 0.1)
Min.Cov.Determinant 64.5 ( 0.1) 64.4 ( 0.5) 63.2 ( 0.1)
Mixture of Gaussians 64.2 ( 0.8) 62.4 ( 1.0) 61.6 ( 1.5)
Naive Parzen 66.9 ( 0.0) 66.9 ( 0.0) 63.4 ( 0.2)
Parzen 56.2 ( 0.2) 57.8 ( 0.2) 63.6 ( 0.2)
k-means 60.4 ( 0.9) 65.2 ( 1.4) 59.2 ( 0.8)
1-Nearest Neighbors 55.3 ( 0.3) 55.4 ( 0.3) 58.9 ( 0.3)
k-Nearest Neighbors 55.3 ( 0.3) 55.4 ( 0.3) 58.9 ( 0.3)
knn, opt-AUC 55.3 ( 0.3) 56.0 ( 1.0) 58.9 ( 0.3)
Nearest-neighbor dist 56.2 ( 0.8) 53.9 ( 0.6) 53.0 ( 0.5)
Principal comp. 62.4 ( 0.1) 61.8 ( 0.0) 55.1 ( 7.6)
Self-Organ. Map 61.5 ( 0.9) 60.1 ( 0.6) 57.9 ( 0.7)
Auto-enc network 52.7 ( 0.4) 52.8 ( 0.1) 58.1 ( 1.0)
Spanning Tree 55.3 ( 0.3) 55.4 ( 0.3) 54.8 ( 0.2)
L_1-ball 65.8 ( 0.1) 49.2 ( 4.0) 40.3 ( 0.3)
k-center 56.6 ( 0.8) 55.9 ( 0.8) 54.5 ( 0.6)
Support vector DD 58.3 ( 0.1) 65.4 ( 0.4) 54.1 ( 0.1)
Minimax Prob. DD 60.6 ( 0.1) 60.3 ( 0.3) 53.7 ( 0.2)
LinProg DD 61.6 ( 0.3) 67.8 ( 0.3) 56.4 ( 0.2)

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