Dataset Satellite, soil with vegetation stubble

Basic characteristics Satellite, soil with vegetation stubble

470

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

3965

outlier objects

36

features

Unsupervised PCA Satellite, soil with vegetation stubble

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, soil with vegetation stubble

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

Results Satellite, soil with vegetation stubble

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

625, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 75.3 ( 0.0) 75.5 ( 0.0) 82.0 ( 0.0)
Min.Cov.Determinant 72.3 ( 0.1) 72.3 ( 0.1) 80.6 ( 0.1)
Mixture of Gaussians 82.3 ( 1.1) 82.9 ( 0.5) 87.2 ( 0.4)
Naive Parzen 87.5 ( 0.0) 87.5 ( 0.0) 79.1 ( 0.0)
Parzen 57.3 ( 0.0) 91.9 ( 0.0) 93.4 ( 0.0)
k-means 68.7 (14.5) 85.4 ( 0.8) 84.6 ( 0.3)
1-Nearest Neighbors 59.8 ( 0.0) 91.9 ( 0.0) 93.5 ( 0.0)
k-Nearest Neighbors 59.8 ( 0.0) 91.9 ( 0.0) 93.5 ( 0.0)
knn, opt-AUC 58.6 ( 1.7) 91.9 ( 0.0) 93.5 ( 0.0)
Nearest-neighbor dist 60.7 ( 0.4) 92.9 ( 0.0) 91.2 ( 0.0)
Principal comp. 66.0 ( 0.0) 72.9 ( 0.0) 55.5 ( 0.0)
Self-Organ. Map 84.0 ( 0.2) 91.0 ( 0.4) 84.3 ( 1.2)
Auto-enc network 65.9 ( 2.0) 69.4 ( 0.1) 65.8 ( 1.9)
Spanning Tree 60.7 ( 0.0) 92.2 ( 0.0) 93.8 ( 0.0)
L_1-ball 79.0 ( 0.0) 78.6 ( 0.0) 73.4 ( 0.0)
k-center 54.1 ( 1.5) 87.3 ( 2.3) 84.0 ( 4.3)
Support vector DD 0.0 ( 0.0) 55.6 ( 0.0) 0.2 ( 0.0)
Minimax Prob. DD 56.4 ( 0.0) 91.9 ( 0.0) 80.7 ( 0.0)
LinProg DD 15.4 ( 1.0) 91.2 ( 0.0) 37.0 ( 0.9)

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