Dataset Thyriod subnormal

Basic characteristics Thyriod subnormal

3488

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 subnormal-functioning is used as target class. Download mat-file with Prtools dataset.

284

outlier objects

21

features

Unsupervised PCA Thyriod subnormal

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 subnormal

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

Results Thyriod subnormal

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

594, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 54.8 ( 0.0) 69.6 ( 0.0) 44.2 ( 0.0)
Min.Cov.Determinant NaN ( 0.0) 63.7 ( 8.7) 79.0 ( 0.1)
Mixture of Gaussians 51.1 ( 3.2) 81.5 ( 1.0) 46.3 ( 3.1)
Naive Parzen 84.4 ( 0.0) 84.4 ( 0.0) 44.5 ( 0.0)
Parzen 54.7 ( 0.0) 87.4 ( 0.0) 44.0 ( 0.0)
k-means 49.5 ( 0.7) 71.9 ( 0.7) 46.9 ( 1.5)
1-Nearest Neighbors 65.0 ( 0.0) 88.5 ( 0.0) 52.8 ( 0.0)
k-Nearest Neighbors 65.0 ( 0.0) 88.5 ( 0.0) 52.8 ( 0.0)
knn, opt-AUC 65.0 ( 0.0) 88.5 ( 0.0) 52.8 ( 0.0)
Nearest-neighbor dist 71.3 ( 0.0) 75.5 ( 0.0) 57.4 ( 0.0)
Principal comp. 54.1 ( 0.0) 70.8 ( 0.0) 42.2 ( 0.0)
Self-Organ. Map 50.5 ( 2.1) 74.7 ( 0.1) 47.0 ( 2.8)
Auto-enc network 49.6 ( 0.0) 70.3 ( 0.6) 46.5 ( 0.0)
Spanning Tree 66.5 ( 0.0) NaN ( 0.0) 46.7 ( 0.0)
L_1-ball 68.8 ( 0.0) 48.9 ( 0.0) 49.2 ( 0.0)
k-center 47.7 ( 1.9) 49.9 ( 1.9) 50.7 ( 1.7)
Support vector DD 50.0 ( 1.2) 50.3 ( 0.0) 50.5 ( 0.0)
Minimax Prob. DD 60.2 ( 0.0) 88.9 ( 0.0) 59.9 ( 0.0)
LinProg DD 51.8 ( 0.0) 50.8 ( 0.1) 49.3 ( 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