Dataset Thyriod normal

Basic characteristics Thyriod normal

93

target objects

One of the thyriod disease databases from UCI containing three classes and about 3800 instances. A patient should be classified as normal, hyper-functioning or subnormal functioning. Class normal is used as target class. Download mat-file with Prtools dataset.

3679

outlier objects

21

features

Unsupervised PCA Thyriod normal

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 Thyriod normal

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

Results Thyriod normal

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

592, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 78.6 ( 0.0) 84.3 ( 0.0) 44.4 ( 0.0)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 77.5 ( 0.0)
Mixture of Gaussians 63.5 ( 2.8) 84.7 ( 4.4) 45.9 ( 3.6)
Naive Parzen 96.1 ( 0.0) 96.1 ( 0.0) 44.5 ( 0.0)
Parzen 63.4 ( 0.0) 90.6 ( 0.0) 48.3 ( 0.0)
k-means 59.7 ( 3.6) 84.4 ( 1.3) 46.3 ( 4.8)
1-Nearest Neighbors 68.3 ( 0.0) 90.6 ( 0.0) 48.6 ( 0.0)
k-Nearest Neighbors 68.3 ( 0.0) 90.6 ( 0.0) 48.6 ( 0.0)
knn, opt-AUC 66.3 ( 1.8) 90.6 ( 0.0) 48.6 ( 0.0)
Nearest-neighbor dist 67.8 ( 0.0) 85.5 ( 0.0) 55.1 ( 0.0)
Principal comp. 68.5 ( 0.0) 67.2 ( 0.0) 45.2 ( 0.0)
Self-Organ. Map 68.8 ( 1.0) 93.1 ( 0.2) 46.2 ( 1.1)
Auto-enc network 68.7 ( 1.9) 78.8 ( 0.7) 45.0 ( 1.3)
Spanning Tree 71.6 ( 0.0) 91.6 ( 0.0) 51.0 ( 0.0)
L_1-ball 51.9 ( 0.0) 51.9 ( 0.0) 54.8 ( 0.0)
k-center 51.9 ( 3.1) 53.3 ( 3.0) 52.5 ( 2.9)
Support vector DD 49.4 ( 2.6) 56.0 ( 0.0) 51.3 ( 0.0)
Minimax Prob. DD 78.1 ( 0.0) 90.9 ( 0.0) 55.2 ( 0.0)
LinProg DD 51.4 ( 0.0) 61.3 ( 0.0) 51.0 ( 0.0)

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