Dataset Heart Cleveland (targetcl. absent)

Basic characteristics Heart Cleveland (targetcl. absent)

164

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 Missing values have been replaced by -1.000000.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.

139

outlier objects

13

features

Unsupervised PCA Heart Cleveland (targetcl. absent)

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. absent)

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

Results Heart Cleveland (targetcl. absent)

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

507, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 75.7 (0.5) 80.0 (0.7) 61.1 (1.0)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 61.2 (0.7)
Mixture of Gaussians 66.5 (0.5) 80.3 (0.9) 62.2 (0.5)
Naive Parzen 82.5 (0.6) 82.5 (0.7) 62.6 (0.7)
Parzen 61.6 (1.0) 79.3 (0.4) 63.4 (0.5)
k-means 59.6 (1.0) 79.3 (0.9) 57.3 (2.4)
1-Nearest Neighbors 61.5 (1.3) 78.0 (0.7) 55.3 (0.7)
k-Nearest Neighbors 61.5 (1.3) 78.0 (0.7) 55.3 (0.7)
Nearest-neighbor dist 56.2 (2.4) 54.4 (2.6) 53.1 (1.1)
Principal comp. 58.0 (1.0) 65.2 (1.1) 52.1 (0.8)
Self-Organ. Map 61.4 (0.7) 79.6 (0.8) 60.1 (0.6)
Auto-enc network 66.0 (4.6) 73.2 (0.8) 58.8 (0.4)
MST 62.5 (1.2) 77.5 (0.5) 55.1 (1.3)
L_1-ball 42.6 (4.6) 73.3 (1.7) 60.2 (0.6)
k-center 61.1 (3.0) 79.3 (2.3) 58.4 (2.5)
Support vector DD 60.9 (14.0) 78.4 (0.6) 56.1 (11.4)
Minimax Prob. DD 61.3 (1.2) 78.3 (0.5) 55.3 (0.7)
LinProg DD 61.5 (1.0) 82.0 (0.6) 55.5 (0.9)
Lof DD 54.6 (2.0) 58.0 (1.0) 50.6 (2.1)
Lof range DD 60.5 (1.3) 65.9 (0.7) 58.6 (0.9)
Loci DD 62.1 (1.2) 81.3 (0.9) 60.2 (1.2)

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