Table 5.
Experimental results for Testing_1 dataset altered with sensor fault anomalies. In bold and italics are reported the best results for the Sparse U-Net and the baseline, respectively.
| Neural Model | Threshold | Accuracy | Precision | Recall | F-Measure |
|---|---|---|---|---|---|
| Deep Autoencoder ( baseline ) | 98th percentile | 0.969 | 0.851 | 1.000 | 0.919 |
| max value | 0.903 | 1.000 | 0.458 | 0.628 | |
| max + tolerance | 0.885 | 1.000 | 0.352 | 0.521 | |
| Sparse U-Net ( Proposed Model ) | 98th percentile | 0.948 | 0.850 | 0.859 | 0.854 |
| max value | 0.940 | 0.985 | 0.675 | 0.801 | |
| max + tolerance | 0.893 | 1.000 | 0.399 | 0.570 |