Dataset Satellite, red soil

Basic characteristics Satellite, red soil

1072

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 red soil is used as target class. Download mat-file with Prtools dataset.

3363

outlier objects

36

features

Unsupervised PCA Satellite, red 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, red soil

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

Results Satellite, red soil

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

621, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 97.2 ( 0.0) 97.2 ( 0.0) 98.8 ( 0.0)
Min.Cov.Determinant 97.3 ( 0.0) 97.3 ( 0.0) 98.6 ( 0.0)
Mixture of Gaussians 84.1 (13.4) 94.2 (10.8) 99.3 ( 0.1)
Naive Parzen 97.9 ( 0.0) 97.9 ( 0.0) 97.4 ( 0.0)
Parzen 76.8 ( 0.0) 79.2 ( 0.0) 99.6 ( 0.0)
k-means 98.7 ( 0.1) 99.0 ( 0.2) 98.7 ( 0.1)
1-Nearest Neighbors 77.5 ( 0.0) 99.8 ( 0.0) 99.6 ( 0.0)
k-Nearest Neighbors 77.5 ( 0.0) 99.8 ( 0.0) 99.6 ( 0.0)
knn, opt-AUC 76.9 ( 0.7) 99.8 ( 0.0) 99.6 ( 0.0)
Nearest-neighbor dist 74.0 ( 0.2) 95.5 ( 0.0) 93.7 ( 0.0)
Principal comp. 89.9 ( 0.0) 90.5 ( 0.0) 99.2 ( 0.0)
Self-Organ. Map 99.0 ( 0.1) 99.3 ( 0.1) 99.0 ( 0.0)
Auto-enc network 90.7 ( 0.1) 89.6 ( 0.0) 94.1 ( 5.9)
Spanning Tree 77.8 ( 0.0) 99.7 ( 0.0) 99.6 ( 0.0)
L_1-ball 96.0 ( 0.0) 94.2 ( 0.0) 74.6 ( 0.0)
k-center 76.1 ( 1.6) 98.9 ( 0.3) 98.3 ( 0.7)
Support vector DD 0.0 ( 0.0) 75.0 ( 0.0) 1.3 ( 0.0)
Minimax Prob. DD 83.3 ( 0.0) 99.8 ( 0.0) 97.4 ( 0.0)
LinProg DD 48.2 ( 0.8) 99.6 ( 0.0) 57.1 ( 0.3)

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