Dataset Ball-bearing

Basic characteristics Ball-bearing

913

target objects

Ball bearing train, downloaded from http://www.sidanet.org/. Task is to distinguish new ball-bearning from bearings where outer race is completely broken, broken cage, damaged case and worn bearnings. Acceleration time signals are measured, signal is FFT-ed resulting in 32 features. Download mat-file with Prtools dataset.

0

outlier objects

32

features

Unsupervised PCA Ball-bearing

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 Ball-bearing

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

Results Ball-bearing

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

510, 0 outliers, AUC (x100) 5x strat. 10-fold
Classifiers Preproc
none unit var PCA 95\%
Gauss 99.9 (0.0) 100.0 (0.0) 86.2 (0.0)
Min.Cov.Determinant 100.0 (0.0) 100.0 (0.0) 86.1 (0.0)
Mixture of Gaussians 100.0 (0.0) 100.0 (0.0) 87.6 (0.2)
Naive Parzen 100.0 (0.0) 100.0 (0.0) 86.7 (0.0)
Parzen 98.7 (0.0) 100.0 (0.0) 86.4 (0.0)
k-means 95.0 (0.2) 100.0 (0.0) 82.5 (0.5)
1-Nearest Neighbors 98.7 (0.0) 100.0 (0.0) 86.7 (0.0)
k-Nearest Neighbors 98.7 (0.0) 100.0 (0.0) 86.7 (0.0)
Nearest-neighbor dist 97.9 (0.0) 99.8 (0.0) 79.0 (0.0)
Principal comp. 99.5 (0.0) 99.9 (0.0) 86.2 (0.0)
Self-Organ. Map 93.2 (0.7) 100.0 (0.0) 80.1 (1.4)
Auto-enc network 99.2 (0.0) 100.0 (0.0) 84.4 (1.3)
MST NaN (0.0) NaN (0.0) NaN (0.0)
L_1-ball 99.9 (0.0) 98.4 (0.0) 86.9 (0.0)
k-center 96.1 (0.3) 100.0 (0.0) 80.5 (0.5)
Support vector DD 0.0 (0.0) 98.7 (0.0) 0.0 (0.0)
Minimax Prob. DD 1.6 (0.0) 100.0 (0.0) 9.2 (0.0)
LinProg DD 5.2 (1.2) 100.0 (0.0) 17.6 (1.8)
Lof DD 99.4 (0.0) 99.9 (0.0) 84.4 (0.0)
Lof range DD 99.2 (0.0) 100.0 (0.0) 86.6 (0.0)
Loci DD 93.9 (0.0) 100.0 (0.0) 81.5 (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