Dataset Pageblocks text

Basic characteristics Pageblocks text

4913

target objects

The Page Blocks Classification Database from UCI. The task is to classify all the blocks ofthe page layout of a document that has been detected by a segmentation process. Class text is used as target class. Download mat-file with Prtools dataset.

560

outlier objects

10

features

Unsupervised PCA Pageblocks text

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 Pageblocks text

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

Results Pageblocks text

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

620, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 56.5 ( 4.8) 59.9 ( 5.9) 53.9 ( 1.5)
Min.Cov.Determinant 93.5 ( 0.0) 93.5 ( 0.0) 61.2 ( 0.0)
Mixture of Gaussians 53.9 ( 1.9) 56.5 ( 6.6) 52.9 ( 2.5)
Naive Parzen 56.1 ( 3.9) 56.1 ( 3.9) 50.0 ( 2.3)
Parzen 50.3 ( 0.8) 50.6 ( 5.1) 50.0 ( 2.3)
k-means 53.1 ( 1.5) 58.9 ( 5.5) 51.5 ( 1.4)
1-Nearest Neighbors 50.5 ( 0.4) 52.4 ( 3.0) 50.2 ( 1.1)
k-Nearest Neighbors 50.5 ( 0.4) 52.4 ( 3.0) 50.2 ( 1.1)
knn, opt-AUC NaN ( 0.0) NaN ( 0.0) NaN ( 0.0)
Nearest-neighbor dist 49.8 ( 1.3) 51.9 ( 1.1) 50.1 ( 1.9)
Principal comp. 55.2 ( 2.4) 59.9 ( 4.5) 0.0 ( 0.0)
Self-Organ. Map 54.0 ( 1.8) 95.9 ( 0.1) 66.7 ( 1.0)
Auto-enc network 53.6 ( 2.2) 59.3 ( 5.1) 51.9 ( 1.4)
Spanning Tree NaN ( 0.0) NaN ( 0.0) NaN ( 0.0)
L_1-ball 90.7 ( 0.0) 91.7 ( 0.1) 67.0 ( 0.1)
k-center 55.7 ( 2.2) 55.9 ( 3.7) 53.6 ( 1.5)
Support vector DD 19.5 ( 0.2) 50.1 ( 5.6) NaN ( 0.0)
Minimax Prob. DD 50.0 ( 0.1) 90.1 ( 0.5) 47.9 ( 0.9)
LinProg DD NaN ( 0.0) NaN ( 0.0) NaN ( 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