Dataset Spectf 0

Basic characteristics Spectf 0

95

target objects

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

254

outlier objects

44

features

Unsupervised PCA Spectf 0

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 0

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

Results Spectf 0

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

556, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 93.4 (0.9) 93.4 (0.8) 83.0 (1.2)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 81.9 (0.8)
Mixture of Gaussians 95.1 (0.8) 94.8 (0.8) 92.2 (1.3)
Naive Parzen 90.7 (1.5) 90.6 (1.5) 89.2 (0.9)
Parzen 95.7 (1.0) 95.7 (1.2) 95.8 (1.1)
k-means 86.0 (0.3) 86.7 (1.0) 82.4 (0.4)
1-Nearest Neighbors 95.7 (1.1) 95.7 (1.2) 95.9 (1.1)
k-Nearest Neighbors 95.7 (1.1) 95.7 (1.2) 95.9 (1.1)
Nearest-neighbor dist 94.4 (0.7) 95.2 (1.1) 94.3 (0.8)
Principal comp. 92.0 (0.7) 91.7 (1.0) 80.7 (2.0)
Self-Organ. Map 89.1 (0.7) 93.0 (0.5) 86.0 (0.3)
Auto-enc network 86.4 (1.4) 90.4 (0.8) 85.0 (1.1)
MST 95.8 (1.0) 95.7 (1.1) 95.8 (1.1)
L_1-ball 81.5 (1.3) 80.3 (0.9) 80.2 (3.3)
k-center 84.2 (1.5) 84.6 (0.6) 81.7 (1.3)
Support vector DD 89.7 (3.2) NaN (0.0) 89.7 (3.2)
Minimax Prob. DD 78.6 (2.9) 95.7 (1.2) 90.1 (2.8)
LinProg DD 94.9 (0.6) 95.6 (0.9) 95.6 (1.0)
Lof DD 84.1 (1.1) 85.1 (1.4) 79.4 (1.0)
Lof range DD 84.9 (0.8) 85.2 (1.1) 81.2 (1.1)
Loci DD 89.7 (0.5) 90.1 (0.4) 87.8 (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