Dataset Delft pump 5x3

Basic characteristics Delft pump 5x3

376

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. From the time signals an 64D envelope spectrum is derived. The 5 measurements are used as independent objects. All normal situations are used in the target class. Download mat-file with Prtools dataset.

1124

outlier objects

64

features

Unsupervised PCA Delft pump 5x3

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 5x3

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

Results Delft pump 5x3

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

542, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 96.0 (0.0) 96.0 (0.0) 58.8 (0.0)
Min.Cov.Determinant NaN (0.0) NaN (0.0) 62.0 (0.0)
Mixture of Gaussians 96.5 (0.4) 94.2 (1.1) 76.4 (3.4)
Naive Parzen 75.9 (0.0) 75.9 (0.0) 69.0 (0.0)
Parzen 94.6 (0.0) 91.4 (0.0) 91.9 (0.0)
k-means 74.6 (0.3) 74.9 (1.2) 57.1 (0.8)
1-Nearest Neighbors 94.6 (0.0) 91.4 (0.0) 92.0 (0.0)
k-Nearest Neighbors 94.6 (0.0) 91.4 (0.0) 92.0 (0.0)
Nearest-neighbor dist 97.6 (0.0) 96.3 (0.0) 93.2 (0.0)
Principal comp. 92.5 (0.0) 96.9 (0.0) 56.1 (0.0)
Self-Organ. Map 87.6 (2.4) 91.8 (1.4) 74.5 (2.5)
Auto-enc network 84.5 (0.0) 80.4 (0.2) 55.1 (0.0)
MST 95.8 (0.0) 93.2 (0.0) 92.9 (0.0)
L_1-ball 56.1 (0.0) 59.7 (0.0) 64.2 (0.0)
k-center 77.9 (1.0) 71.9 (1.9) 68.8 (1.9)
Support vector DD 94.3 (0.1) 71.3 (0.4) 92.8 (0.6)
Minimax Prob. DD 94.5 (0.0) 91.4 (0.0) 93.4 (0.0)
LinProg DD 95.1 (0.0) 81.9 (0.0) 88.9 (0.0)
Lof DD 98.6 (0.0) 97.8 (0.0) 94.6 (0.0)
Lof range DD 97.4 (0.0) 96.2 (0.0) 91.9 (0.0)
Loci DD 92.3 (0.0) 83.0 (0.0) 84.3 (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