Dataset Ionosphere good

Basic characteristics Ionosphere good

225

target objects

The Ionosphere database from UCI, to classify radar returns from the ionosphere. The target class contains the radar returns that show evidence for some structure in the ionosphere. Download mat-file with Prtools dataset.

126

outlier objects

34

features

Unsupervised PCA Ionosphere good

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 Ionosphere good

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

Results Ionosphere good

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

588, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 96.3 ( 0.3) 96.3 ( 0.3) 92.5 ( 0.6)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 89.2 ( 1.1)
Mixture of Gaussians 96.6 ( 0.4) 96.4 ( 0.3) 93.7 ( 0.8)
Naive Parzen 93.2 ( 0.3) 93.2 ( 0.3) 94.2 ( 0.6)
Parzen 95.6 ( 0.4) 96.8 ( 0.4) 92.5 ( 0.8)
k-means 97.0 ( 0.1) 96.8 ( 0.6) 90.1 ( 1.0)
1-Nearest Neighbors 95.5 ( 0.4) 96.8 ( 0.3) 91.4 ( 0.7)
k-Nearest Neighbors 95.5 ( 0.4) 96.8 ( 0.3) 91.4 ( 0.7)
knn, opt-AUC 95.5 ( 0.4) 96.8 ( 0.3) 91.8 ( 0.7)
Nearest-neighbor dist 86.4 ( 0.3) 86.3 ( 0.6) 75.7 ( 0.4)
Principal comp. 97.8 ( 0.2) 94.7 ( 0.2) 94.3 ( 0.7)
Self-Organ. Map 94.7 ( 0.5) 96.3 ( 0.4) 87.8 ( 1.2)
Auto-enc network 95.2 ( 0.3) 94.7 ( 0.2) 87.2 ( 0.7)
Spanning Tree 95.5 ( 0.4) 96.8 ( 0.3) 91.1 ( 0.6)
L_1-ball 65.8 (11.7) 65.8 (11.7) 94.0 ( 0.5)
k-center 94.0 ( 0.6) 94.9 ( 0.4) 85.8 ( 1.8)
Support vector DD 96.6 ( 0.4) 49.1 ( 6.8) 93.7 ( 0.9)
Minimax Prob. DD 96.0 ( 0.4) 96.9 ( 0.3) 93.1 ( 0.9)
LinProg DD 96.2 ( 0.4) 97.1 ( 0.3) 90.0 ( 0.7)

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