Dataset Delft pump AR app.

Basic characteristics Delft pump AR app.

189

target objects

The Delft pump dataset, with some description in A. Ypma, Learning methods for machine vibration analysis and health monitoring, thesis Delft university of Technology, 2001. From a submersible pump 5 vibration measurements are taken under different normal and abnormal conditions. On the time signals an AR model is fitted, giving 32 features. The 5 measurements are combined to one object, giving an 160D feature space. Download mat-file with Prtools dataset.

531

outlier objects

160

features

Unsupervised PCA Delft pump AR app.

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 Delft pump AR app.

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

Results Delft pump AR app.

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

541, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 100.0 (0.0) 100.0 (0.0) 97.9 (0.0)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 96.7 (0.2)
Mixture of Gaussians 99.9 (0.0) 99.9 (0.1) 99.0 (0.2)
Naive Parzen 97.9 (0.0) 97.8 (0.0) 98.0 (0.0)
Parzen 99.9 (0.0) 99.8 (0.0) 99.3 (0.0)
k-means 98.9 (0.3) 99.1 (0.2) 95.1 (1.2)
1-Nearest Neighbors 99.8 (0.0) 99.8 (0.0) 99.3 (0.0)
k-Nearest Neighbors 99.8 (0.0) 99.8 (0.0) 99.3 (0.0)
Nearest-neighbor dist 99.7 (0.0) 99.7 (0.0) 98.8 (0.0)
Principal comp. 100.0 (0.0) 99.9 (0.0) 97.4 (0.0)
Self-Organ. Map 99.7 (0.1) 99.5 (0.1) 98.4 (0.1)
Auto-enc network 99.1 (0.1) 87.8 (17.2) 95.4 (0.0)
MST 99.9 (0.0) 99.8 (0.0) 99.4 (0.0)
L_1-ball 93.6 (0.0) 90.9 (0.0) 93.5 (0.0)
k-center 99.1 (0.5) 99.2 (0.2) 96.6 (0.7)
Support vector DD 97.9 (0.1) 50.4 (0.1) 88.6 (0.1)
Minimax Prob. DD 98.9 (0.0) 99.8 (0.0) 92.4 (0.0)
LinProg DD 95.3 (0.0) 99.6 (0.0) 83.3 (0.0)
Lof DD 99.8 (0.0) 99.8 (0.0) 98.1 (0.0)
Lof range DD 99.8 (0.0) 99.8 (0.0) 97.5 (0.0)
Loci DD 96.6 (0.0) 97.0 (0.0) 91.9 (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