Dataset Spambase other

Basic characteristics Spambase other

2788

target objects

Spambase database from UCI. The spam concept is very diverse: advertisements for products/web sites, make money fast schemes, chain letters, pornography... This class is used as target. Download mat-file with Prtools dataset.

1813

outlier objects

57

features

Unsupervised PCA Spambase other

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 Spambase other

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

Results Spambase other

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

553, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 64.7 ( 0.1) 79.6 ( 0.1) 50.7 ( 0.0)
Min.Cov.Determinant NaN ( 0.0) NaN ( 0.0) 68.0 ( 0.0)
Mixture of Gaussians 73.3 ( 0.1) 83.0 ( 0.4) 62.8 ( 0.2)
Naive Parzen 78.6 ( 0.2) 78.6 ( 0.2) 73.8 ( 0.1)
Parzen 78.9 ( 0.1) 88.1 ( 0.1) 73.8 ( 0.1)
k-means 69.4 ( 0.1) 80.2 ( 0.2) 64.7 ( 0.1)
1-Nearest Neighbors 80.6 ( 0.1) 88.3 ( 0.1) 75.6 ( 0.1)
k-Nearest Neighbors 80.6 ( 0.1) 88.3 ( 0.1) 75.6 ( 0.1)
Nearest-neighbor dist 62.6 ( 0.5) 70.0 ( 0.5) 58.6 ( 0.4)
Principal comp. 64.9 ( 0.1) 68.9 ( 0.3) 51.0 (10.6)
Self-Organ. Map 54.1 ( 2.3) 80.7 ( 0.1) 51.0 ( 1.5)
Auto-enc network NaN ( 0.0) NaN ( 0.0) 63.8 ( 1.6)
MST 80.2 ( 0.1) 88.2 ( 0.1) 50.1 ( 0.0)
L_1-ball 84.5 ( 0.3) 77.4 ( 0.3) 50.7 ( 0.0)
k-center 65.7 ( 1.2) 76.9 ( 0.9) 49.6 ( 0.7)
Support vector DD NaN ( 0.0) NaN ( 0.0) NaN ( 0.0)
Minimax Prob. DD 80.4 ( 0.1) 88.2 ( 0.1) 63.7 ( 0.5)
LinProg DD 80.8 ( 0.1) 83.2 ( 0.1) 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