Dataset Spambase spam

Basic characteristics Spambase spam

1813

target objects

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

2788

outlier objects

57

features

Unsupervised PCA Spambase spam

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 spam

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

Results Spambase spam

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

552, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 63.2 (0.0) 84.8 (0.0) 61.9 (0.1)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 51.0 (0.2)
Mixture of Gaussians 46.6 (0.5) 85.5 (0.2) 56.2 (0.2)
Naive Parzen 76.8 (0.3) 76.8 (0.3) 46.4 (0.4)
Parzen 37.8 (0.4) 89.0 (0.1) 48.1 (0.3)
k-means 48.8 (0.2) 83.9 (0.1) 48.9 (0.3)
1-Nearest Neighbors 66.5 (0.3) 89.8 (0.0) 54.1 (0.4)
k-Nearest Neighbors 66.5 (0.3) 89.8 (0.0) 54.1 (0.4)
Nearest-neighbor dist 71.3 (0.4) 73.6 (0.5) 65.3 (0.2)
Principal comp. 41.5 (1.8) 86.0 (0.1) 37.8 (1.2)
Self-Organ. Map 63.3 (2.1) 84.1 (0.1) 62.8 (0.8)
Auto-enc network 65.6 (4.3) 83.5 (0.1) 47.9 (3.4)
MST 68.1 (0.4) 89.8 (0.0) 55.9 (0.3)
L_1-ball 74.7 (2.9) 75.9 (0.5) 38.7 (0.8)
k-center 56.2 (1.2) 84.6 (0.4) 53.0 (0.9)
Support vector DD NaN (0.0) NaN (0.0) NaN (0.0)
Minimax Prob. DD 66.4 (0.3) 89.9 (0.1) 50.6 (0.3)
LinProg DD 64.6 (0.3) 86.4 (0.1) 52.1 (0.5)
Lof DD 61.9 (0.3) 77.6 (0.3) 53.3 (0.5)
Lof range DD 55.4 (0.2) 78.1 (0.3) 53.5 (0.2)
Loci DD 62.5 (0.1) 86.1 (0.1) 58.0 (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