Dataset Iris (targetcl. Iris-virginica)

Basic characteristics Iris (targetcl. Iris-virginica)

50

target objects

Iris Plant database from UCI. A classic dataset in the pattern recognition literature. The original dataset is a multiclass classification problem, introduced by R.A. Fisher, The use of multiple measurements in taxonomic problems. Ann Eugenics, 7:179--188, 1936. Download mat-file with Prtools dataset.

100

outlier objects

4

features

Unsupervised PCA Iris (targetcl. Iris-virginica)

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 Iris (targetcl. Iris-virginica)

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

Results Iris (targetcl. Iris-virginica)

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

503, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 97.8 (0.6) 97.8 (0.5) 98.6 (0.5)
Min.Cov.Determinant 97.6 (0.2) 97.6 (0.2) 98.3 (0.3)
Mixture of Gaussians 97.1 (0.7) 97.4 (0.9) 97.7 (0.4)
Naive Parzen 95.4 (1.1) 95.4 (1.1) 98.1 (0.7)
Parzen 96.5 (0.8) 96.8 (0.9) 96.8 (0.7)
k-means 95.4 (0.5) 95.8 (0.9) 96.2 (0.6)
1-Nearest Neighbors 97.0 (0.8) 96.7 (0.9) 97.2 (0.7)
k-Nearest Neighbors 97.0 (0.8) 96.7 (0.9) 97.2 (0.7)
Nearest-neighbor dist 91.8 (1.6) 87.8 (2.2) 91.2 (0.9)
Principal comp. 90.9 (4.7) 90.4 (1.3) 98.3 (0.4)
Self-Organ. Map 96.6 (0.3) 96.9 (0.9) 97.7 (0.5)
Auto-enc network 95.6 (1.4) 96.1 (0.7) 97.0 (2.1)
MST 97.0 (0.7) 96.7 (1.0) 97.3 (0.5)
L_1-ball 99.0 (0.2) 98.3 (0.4) 93.4 (0.9)
k-center 96.9 (0.7) 96.0 (0.9) 96.9 (0.8)
Support vector DD 98.1 (0.8) 97.3 (0.4) 98.3 (0.7)
Minimax Prob. DD 97.2 (0.8) 96.9 (0.7) 97.7 (0.3)
LinProg DD 98.6 (0.4) 98.2 (0.4) 98.6 (0.5)
Lof DD 96.6 (1.2) 96.4 (0.6) 97.4 (0.9)
Lof range DD 96.0 (0.9) 96.6 (0.6) 97.0 (1.2)
Loci DD 95.3 (1.3) 95.4 (0.5) 96.1 (0.6)

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