Dataset Ionosphere bad

Basic characteristics Ionosphere bad

126

target objects

The Ionosphere database from UCI, to classify radar returns from the ionosphere. The target class contains the radar returns that show no evidence for some structure in the ionosphere. Download mat-file with Prtools dataset.

225

outlier objects

34

features

Unsupervised PCA Ionosphere bad

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 bad

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

Results Ionosphere bad

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

589, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 24.6 ( 0.4) 24.6 ( 0.4) 27.7 ( 0.5)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 25.4 ( 0.7)
Mixture of Gaussians 23.3 ( 0.6) 23.3 ( 0.2) 25.0 ( 1.2)
Naive Parzen 81.3 ( 0.6) 81.3 ( 0.6) 27.6 ( 0.7)
Parzen 26.9 ( 0.5) 26.9 ( 0.4) 28.9 ( 0.5)
k-means 24.1 ( 1.0) 24.1 ( 0.7) 25.1 ( 0.8)
1-Nearest Neighbors 25.0 ( 0.7) 25.1 ( 0.6) 25.5 ( 0.7)
k-Nearest Neighbors 25.0 ( 0.7) 25.1 ( 0.6) 25.5 ( 0.7)
knn, opt-AUC 26.6 ( 0.8) 26.0 ( 1.0) 28.5 ( 0.8)
Nearest-neighbor dist 30.1 ( 1.2) 30.0 ( 1.0) 31.7 ( 1.2)
Principal comp. 23.9 ( 0.9) 23.5 ( 0.6) 28.5 ( 0.8)
Self-Organ. Map 25.2 ( 0.9) 24.5 ( 0.6) 25.8 ( 0.4)
Auto-enc network 23.3 ( 0.7) 23.1 ( 0.6) 24.2 ( 0.5)
Spanning Tree 25.4 ( 0.6) 25.3 ( 0.7) 24.8 ( 0.6)
L_1-ball 48.9 (16.7) 48.9 (16.7) 28.6 ( 1.8)
k-center 24.9 ( 1.1) 26.2 ( 1.3) 26.7 ( 2.0)
Support vector DD 25.4 ( 1.1) 46.9 ( 6.0) 24.2 ( 0.5)
Minimax Prob. DD 24.5 ( 0.5) 24.9 ( 0.7) 24.6 ( 0.6)
LinProg DD 34.9 ( 0.8) 29.7 ( 0.4) 37.4 ( 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