Dataset Leukemia 2

Basic characteristics Leukemia 2

47

target objects

The Leukemia gene expression dataset, presented in T.R.Golub et al., Molecular Classification of Cancer: class discovery and class prediction by gene expression monitoring, Science, 1999. Download mat-file with Prtools dataset.

25

outlier objects

3571

features

Unsupervised PCA Leukemia 2

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 Leukemia 2

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

Results Leukemia 2

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

573, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 92.1 (1.8) 86.5 (3.2) 85.4 (3.7)
Min.Cov.Determinant NaN (0.0) NaN (0.0) NaN (0.0)
Mixture of Gaussians NaN (0.0) NaN (0.0) NaN (0.0)
Naive Parzen 90.2 (4.4) 90.2 (4.4) 80.2 (3.3)
Parzen NaN (0.0) 58.9 (30.2) 90.8 (1.6)
k-means 92.0 (1.2) 86.7 (2.6) 82.8 (3.7)
1-Nearest Neighbors 92.3 (2.0) 87.3 (2.8) 88.4 (1.6)
k-Nearest Neighbors 92.3 (2.0) 87.3 (2.8) 88.4 (1.6)
Nearest-neighbor dist 77.7 (2.2) 84.2 (4.6) 58.5 (2.5)
Principal comp. 92.2 (1.9) 86.5 (3.2) 77.4 (5.0)
Self-Organ. Map 93.6 (1.5) 86.6 (2.5) 85.1 (4.2)
Auto-enc network NaN (0.0) NaN (0.0) 71.3 (6.7)
MST 90.6 (1.3) 86.8 (2.6) 87.6 (3.1)
L_1-ball 60.2 (28.0) 60.2 (28.0) 73.6 (4.4)
k-center 91.1 (2.4) 87.1 (2.7) 81.6 (6.2)
Support vector DD 58.9 (30.2) 58.9 (30.2) 58.9 (30.2)
Minimax Prob. DD 92.3 (2.0) 0.0 (0.0) 88.4 (1.6)
LinProg DD 94.6 (1.2) 58.9 (30.2) 91.5 (2.1)
Lof DD 90.3 (1.6) 86.4 (3.2) 42.8 (4.6)
Lof range DD 91.4 (1.4) 86.4 (3.2) 42.7 (6.3)
Loci DD 91.0 (3.1) 85.9 (3.5) 69.7 (6.1)

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