Dataset Spectf 1

Basic characteristics Spectf 1

254

target objects

SPECTF heart database from UCI. Describes diagnosing of cardiac Single Proton Emission Computed Tomography (SPECT) images. Abnormal patients are used as target class. Download mat-file with Prtools dataset.

95

outlier objects

44

features

Unsupervised PCA Spectf 1

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 Spectf 1

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

Results Spectf 1

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

557, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 28.4 (0.5) 28.6 (0.5) 24.0 (0.7)
Min.Cov.Determinant 29.9 (0.6) 29.9 (0.4) 25.0 (0.9)
Mixture of Gaussians 27.9 (1.3) 30.2 (0.6) 24.1 (1.1)
Naive Parzen 26.0 (0.7) 26.0 (0.7) 23.6 (1.0)
Parzen 44.5 (0.5) 46.6 (0.1) 43.4 (0.9)
k-means 24.2 (1.4) 25.5 (1.2) 24.9 (0.5)
1-Nearest Neighbors 45.0 (0.4) 46.9 (0.4) 45.3 (0.7)
k-Nearest Neighbors 45.0 (0.4) 46.9 (0.4) 45.3 (0.7)
Nearest-neighbor dist 58.0 (1.6) 56.8 (0.4) 58.7 (1.5)
Principal comp. 28.9 (0.6) 29.3 (0.7) 29.1 (1.4)
Self-Organ. Map 24.6 (0.8) 29.8 (1.3) 25.1 (1.1)
Auto-enc network 26.7 (2.1) 26.2 (1.0) 24.9 (1.2)
MST 45.8 (0.5) 47.3 (0.6) 46.4 (0.4)
L_1-ball 37.7 (1.1) 39.6 (0.9) 31.6 (2.8)
k-center 29.1 (3.9) 30.0 (2.2) 29.4 (3.0)
Support vector DD 57.1 (11.1) NaN (0.0) 57.1 (11.1)
Minimax Prob. DD 34.0 (0.7) 46.8 (0.2) 38.8 (1.0)
LinProg DD 46.4 (2.8) 35.5 (0.8) 45.4 (1.0)
Lof DD 36.3 (1.1) 36.1 (1.6) 37.4 (1.9)
Lof range DD 26.3 (0.9) 25.8 (1.0) 26.0 (1.2)
Loci DD 37.7 (1.9) 39.6 (0.5) 37.2 (2.1)

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