Dataset Sonar (targetcl. rocks)

Basic characteristics Sonar (targetcl. rocks)

97

target objects

The Sonar dataset from the undocumented databases from UCI. The task is to train a network to discriminate between sonar signals bounced off a metal cylinder and those bounced off a roughly cylindrical rock. Download mat-file with Prtools dataset.

111

outlier objects

60

features

Unsupervised PCA Sonar (targetcl. rocks)

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 Sonar (targetcl. rocks)

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

Results Sonar (targetcl. rocks)

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

509, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 60.3 (0.2) 61.5 (0.8) 54.3 (0.3)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 59.9 (1.0)
Mixture of Gaussians 66.3 (1.0) 67.7 (0.9) 62.9 (2.1)
Naive Parzen 67.0 (1.1) 67.0 (1.1) 55.4 (1.2)
Parzen 68.1 (0.7) 73.0 (0.8) 67.9 (1.3)
k-means 56.0 (1.2) 67.4 (1.0) 56.4 (1.8)
1-Nearest Neighbors 68.2 (0.8) 73.2 (1.0) 68.8 (0.9)
k-Nearest Neighbors 68.2 (0.8) 73.2 (1.0) 68.8 (0.9)
Nearest-neighbor dist 71.6 (2.2) 72.0 (1.9) 70.7 (0.8)
Principal comp. 56.4 (0.6) 63.5 (1.2) 54.3 (1.5)
Self-Organ. Map 65.5 (1.3) 72.6 (0.9) 65.6 (1.7)
Auto-enc network 61.1 (0.4) 66.5 (2.2) 61.2 (0.5)
MST 67.1 (0.6) 73.7 (0.6) 68.7 (0.9)
L_1-ball 59.5 (1.0) 61.8 (1.3) 51.8 (0.6)
k-center 60.0 (2.5) 69.8 (1.2) 56.9 (3.1)
Support vector DD 58.9 (1.9) 68.6 (6.4) 58.8 (1.5)
Minimax Prob. DD 59.0 (1.1) 73.1 (0.9) 60.1 (2.0)
LinProg DD 63.9 (1.0) 69.3 (1.1) 60.3 (1.9)
Lof DD 68.2 (1.4) 73.7 (1.8) 68.9 (1.8)
Lof range DD 66.1 (1.4) 71.3 (1.4) 66.6 (1.5)
Loci DD 62.0 (0.8) 69.7 (0.4) 58.9 (0.8)

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