Dataset Satellite, cotton crop

Basic characteristics Satellite, cotton crop

479

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 cotton crop is used as target class. Download mat-file with Prtools dataset.

3956

outlier objects

36

features

Unsupervised PCA Satellite, cotton crop

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, cotton crop

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

Results Satellite, cotton crop

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

622, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 88.1 ( 0.0) 88.0 ( 0.0) 90.7 ( 0.0)
Min.Cov.Determinant 89.6 ( 0.2) 89.6 ( 0.2) 93.5 ( 0.2)
Mixture of Gaussians 97.4 ( 0.4) 97.5 ( 0.5) 97.5 ( 0.4)
Naive Parzen 98.0 ( 0.0) 98.0 ( 0.0) 93.1 ( 0.0)
Parzen 98.8 ( 0.0) 99.0 ( 0.0) 98.4 ( 0.0)
k-means 84.6 (18.5) 98.3 ( 0.5) 97.8 ( 0.8)
1-Nearest Neighbors 53.5 ( 0.0) 99.0 ( 0.0) 98.3 ( 0.0)
k-Nearest Neighbors 53.5 ( 0.0) 99.0 ( 0.0) 98.3 ( 0.0)
knn, opt-AUC 60.2 ( 3.4) 99.0 ( 0.0) 98.3 ( 0.0)
Nearest-neighbor dist 55.7 ( 0.0) 94.7 ( 0.0) 89.5 ( 0.0)
Principal comp. 76.3 ( 0.0) 78.6 ( 0.0) 58.8 ( 0.0)
Self-Organ. Map 99.1 ( 0.2) 99.0 ( 0.7) 98.7 ( 0.4)
Auto-enc network 77.7 ( 4.3) 78.6 ( 0.0) 59.2 ( 0.4)
Spanning Tree 53.6 ( 0.0) 99.0 ( 0.0) 98.3 ( 0.0)
L_1-ball 94.1 ( 0.0) 99.1 ( 0.0) 54.8 ( 0.0)
k-center 62.1 ( 5.5) 97.5 ( 1.5) 97.9 ( 2.3)
Support vector DD 0.0 ( 0.0) 37.6 ( 0.0) 0.4 ( 0.0)
Minimax Prob. DD 29.5 ( 0.0) 99.1 ( 0.0) 78.5 ( 0.0)
LinProg DD 3.1 ( 0.9) 99.4 ( 0.0) 8.2 ( 1.6)

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