Table 10 CoLog in comparison to other log anomaly detection methods on the BlueGene/L dataset. Bold numbers indicate the outstanding results.
Anomaly detection technique | Precision | Recall | F1-Score | Accuracy |
|---|---|---|---|---|
Supervised Methods | ||||
Logistic Regression143 | 54.028 | 51.852 | 52.092 | 90.368 |
Support Vector Machines109 | 46.314 | 50.000 | 48.087 | 92.628 |
Decision Tree110 | 60.576 | 50.998 | 50.303 | 92.348 |
Attentional BiLSTM63 | 97.640 | 97.955 | 97.792 | 97.902 |
Convolutional Neural Network46 | 97.640 | 97.955 | 97.792 | 97.902 |
pylogsentiment38 | 99.892 | 99.963 | 99.928 | 99.980 |
Unsupervised Methods | ||||
Isolation Forest144 | 53.081 | 50.047 | 51.519 | 47.389 |
Principal Component Analysis12 | 51.168 | 54.260 | 38.970 | 48.487 |
LSTM36 | 97.414 | 98.296 | 97.806 | 97.902 |
Transformer80 | 97.640 | 97.955 | 97.792 | 97.902 |
\(\hbox {CoLog}^{1}\) | \(\mathbf {99.999}\) | \(\mathbf {99.990}\) | \(\mathbf {99.994}\) | \(\mathbf {99.998}\) |
