Dataset Ecoli (targetcl. periplasm)

Basic characteristics Ecoli (targetcl. periplasm)

52

target objects

The Ecoli database from UCI. Goal is to Predict the localization site of protein in a cell, by Kenta Nakai Institue of Molecular and Cellular Biology Osaka, University. Download mat-file with Prtools dataset.

284

outlier objects

7

features

Unsupervised PCA Ecoli (targetcl. periplasm)

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 Ecoli (targetcl. periplasm)

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

Results Ecoli (targetcl. periplasm)

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

519, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 92.9 (0.3) 92.9 (0.3) 92.8 (0.7)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 94.1 (0.4)
Mixture of Gaussians 92.0 (0.4) 92.9 (0.9) 92.0 (1.1)
Naive Parzen 93.0 (0.8) 93.0 (0.8) 91.0 (1.1)
Parzen 92.2 (0.4) 92.9 (0.5) 93.5 (0.6)
k-means 89.1 (1.6) 87.8 (1.5) 88.6 (1.5)
1-Nearest Neighbors 90.2 (0.9) 90.6 (0.8) 89.8 (1.0)
k-Nearest Neighbors 90.2 (0.9) 90.6 (0.8) 89.8 (1.0)
Nearest-neighbor dist 60.8 (1.6) 66.4 (2.0) 59.5 (1.7)
Principal comp. 66.9 (1.1) 65.5 (1.3) 65.7 (0.9)
Self-Organ. Map 89.0 (1.1) 89.8 (0.4) 88.6 (0.8)
Auto-enc network 87.8 (1.0) 88.8 (2.3) 85.8 (2.3)
MST 88.7 (0.9) 89.9 (0.9) 88.4 (0.8)
L_1-ball 39.9 (13.2) 39.9 (13.2) 82.8 (1.8)
k-center 86.3 (1.2) 87.0 (2.3) 86.2 (1.6)
Support vector DD 89.4 (0.8) 92.2 (1.0) 89.0 (0.7)
Minimax Prob. DD 80.2 (0.5) 92.2 (0.7) 79.0 (0.6)
LinProg DD 89.6 (0.5) 94.7 (0.4) 89.2 (0.5)
Lof DD 88.9 (1.1) 87.0 (0.9) 88.2 (1.0)
Lof range DD 95.3 (0.6) 94.7 (0.2) 94.9 (0.6)
Loci DD 94.1 (0.6) 93.9 (0.4) 93.7 (0.7)

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