Dataset Satellite, grey soil

Basic characteristics Satellite, grey soil

961

target objects

The Landsat Satellite database from the Statlog Project. The classification associated with the central pixel in each 3x3 neighbourhood in a satellite image, based on multi-spectral values of pixels. Class grey soil is used as target class. Download mat-file with Prtools dataset.

3474

outlier objects

36

features

Unsupervised PCA Satellite, grey soil

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 Satellite, grey soil

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

Results Satellite, grey soil

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

623, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 91.8 ( 0.0) 91.8 ( 0.0) 93.7 ( 0.0)
Min.Cov.Determinant 90.9 ( 0.0) 90.9 ( 0.0) 92.3 ( 0.0)
Mixture of Gaussians 40.0 ( 0.9) 39.6 ( 0.7) 85.1 (24.2)
Naive Parzen 97.1 ( 0.1) 97.2 ( 0.1) 92.8 ( 0.0)
Parzen 53.5 ( 0.0) 53.5 ( 0.0) 54.0 ( 0.0)
k-means 88.6 (18.6) 96.8 ( 0.6) 96.7 ( 0.6)
1-Nearest Neighbors 55.1 ( 0.0) 97.3 ( 0.0) 97.4 ( 0.0)
k-Nearest Neighbors 55.1 ( 0.0) 97.3 ( 0.0) 97.4 ( 0.0)
knn, opt-AUC 55.0 ( 0.2) 97.3 ( 0.0) 97.4 ( 0.0)
Nearest-neighbor dist 53.9 ( 0.3) 95.2 ( 0.0) 93.0 ( 0.0)
Principal comp. 78.8 ( 0.0) 78.2 ( 0.0) 83.4 ( 0.0)
Self-Organ. Map 96.4 ( 0.2) 97.1 ( 0.3) 96.3 ( 0.2)
Auto-enc network 83.8 ( 1.0) 83.6 ( 0.2) 88.8 ( 5.1)
Spanning Tree 52.8 ( 0.0) 97.3 ( 0.0) 97.5 ( 0.0)
L_1-ball 97.5 ( 0.0) 97.0 ( 0.0) 71.3 ( 0.0)
k-center 53.8 ( 1.1) 97.1 ( 0.4) 97.5 ( 1.0)
Support vector DD 0.0 ( 0.0) 19.5 ( 0.0) 0.0 ( 0.0)
Minimax Prob. DD 88.2 ( 0.0) 97.4 ( 0.0) 92.7 ( 0.0)
LinProg DD 18.2 ( 0.1) 52.8 ( 0.0) 18.4 ( 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