Dataset Heart Cleveland (targetcl. present)

Basic characteristics Heart Cleveland (targetcl. present)

139

target objects

The Cleveland database from the Heart Disease Databases from UCI. The class disease-presence is used as target class. Missing values have been replaced by -1.000000. Download mat-file with Prtools dataset.

164

outlier objects

13

features

Unsupervised PCA Heart Cleveland (targetcl. present)

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 Heart Cleveland (targetcl. present)

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

Results Heart Cleveland (targetcl. present)

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

506, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 60.9 (0.3) 63.8 (0.7) 54.7 (0.6)
Min.Cov.Determinant 66.8 (1.8) 66.7 (1.8) 52.9 (0.5)
Mixture of Gaussians 55.0 (1.6) 66.1 (1.5) 52.9 (0.4)
Naive Parzen 73.2 (0.4) 73.3 (0.5) 50.6 (0.5)
Parzen 49.7 (0.6) 65.6 (0.6) 49.3 (0.5)
k-means 56.1 (0.8) 63.8 (1.1) 55.4 (0.3)
1-Nearest Neighbors 49.2 (0.6) 63.7 (0.8) 47.5 (1.2)
k-Nearest Neighbors 49.2 (0.6) 63.7 (0.8) 47.5 (1.2)
Nearest-neighbor dist 48.9 (1.0) 58.4 (1.1) 51.0 (1.0)
Principal comp. 56.8 (0.9) 53.6 (0.9) 41.1 (10.4)
Self-Organ. Map 54.4 (1.4) 65.6 (1.3) 53.1 (0.4)
Auto-enc network 54.4 (3.9) 50.0 (0.8) 52.3 (1.3)
MST 50.5 (0.8) 64.2 (0.4) 49.1 (1.2)
L_1-ball 66.1 (6.5) 78.9 (0.8) 51.8 (1.7)
k-center 51.1 (4.6) 67.3 (2.6) 52.7 (4.0)
Support vector DD 41.3 (13.0) 64.4 (0.5) 42.6 (11.1)
Minimax Prob. DD 49.0 (0.6) 64.3 (0.5) 47.5 (1.3)
LinProg DD 49.4 (0.4) 69.5 (0.6) 47.3 (1.1)
Lof DD 52.0 (0.8) 51.5 (1.7) 49.1 (1.2)
Lof range DD 51.9 (1.0) 47.5 (1.3) 50.4 (1.6)
Loci DD 52.4 (1.3) 68.8 (1.0) 52.8 (1.0)

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