Table 4 F1-score and balanced accuracy results (%) of the proposed model with LSTM only, LSTM combined with self-attention, against the transformer over all datasets using the first scenario
From: Deep reinforcement learning for data-efficient weakly supervised business process anomaly detection
Datasets | Balanced accuracy | F1-score | ||||
|---|---|---|---|---|---|---|
LSTM | LSTM w/Self Att | Transformer | LSTM | LSTM w/Self Att | Transformer | |
DS1 | 95.67 ± 0.02 | 98.90 ± 0.07 | 94.34 ± 0.02 | 92.64 ± 0.11 | 97.41 ± 0.05 | 92.54 ± 0.15 |
DS2 | 95.46 ± 0.01 | 97.02 ± 0.05 | 93.88 ± 0.05 | 91.90 ± 0.09 | 95.60 ± 0.07 | 91.10 ± 0.04 |
BPIC12 | 85.96 ± 0.01 | 93.37 ± 0.06 | 83.24 ± 0.03 | 82.32 ± 0.04 | 87.56 ± 0.01 | 79.96 ± 0.02 |
BPIC17 | 90.61 ± 0.15 | 95.04 ± 0.04 | 88.76 ± 0.04 | 87.43 ± 0.04 | 91.40 ± 0.05 | 84.23 ± 0.10 |