Research Article
An LSTM-Autoencoder Architecture for Anomaly Detection Applied on Compressors Audio Data
Table 14
LSTM-AE result on test dataset performance with baseline model comparison (created by authors).
| Model architecture | Loss function | Activation function | Optimizer | Accuracy | Precision | Recall | F1 |
| GRU | Mean square error | SoftMax | Adam | 87% | 87% | 52% | 65% | GRU-1 | Sparse categorical cross entropy | Sigmoid | Adam | 99% | 100% | 98% | 99% | GRU-2 | Sparse categorical cross entropy | SoftMax | Adam | 99% | 99% | 99% | 99% | Stacked GRU | Sparse categorical cross entropy | SoftMax | Adam | 99% | 98% | 100% | 99% | LSTM | Sparse categorical cross entropy | Sigmoid | Adam | 97% | 100% | 97% | 98% | LSTM-1 | Sparse categorical cross entropy | SoftMax | Adam | 99% | 100% | 98% | 99% | Stacked LSTM | Sparse categorical cross entropy | SoftMax | Adam | 99% | 100% | 99% | 99% | LSTM- Autoencoder | Sparse categorical cross entropy | SoftMax | Adam | 100% | 100% | 100% | 100% |
|
|