From: Deep reinforcement learning for data-efficient weakly supervised business process anomaly detection
Data | Proposed model | Denoising Autoencoder [21] | Variational autoencoder [22] | BINET [5] | Bi-LSTM VAE [23] | Process mining model | |
---|---|---|---|---|---|---|---|
Datasets | Scenarios | ||||||
DS1 | All | 97.41 ± 0.00 | 83.12 ± 0.18 | 85.09 ± 0.12 | 90.30 ± 0.08 | 91.22 ± 0.08 | 89.00 |
Rework | 96.66 ± 0.07 | 81.58 ± 0.03 | 83.10 ± 0.18 | 88.50 ± 0.05 | 89.30 ± 0.02 | 86.85 | |
Skip | 95.47 ± 0.01 | 79.19 ± 0.06 | 81.34 ± 0.02 | 85.91 ± 0.01 | 86.84 ± 0.06 | 84.10 | |
Switch | 92.51 ± 0.12 | 68.84 ± 0.18 | 70.63 ± 0.02 | 76.74 ± 0.03 | 77.40 ± 0.01 | 75.68 | |
DS2 | All | 95.60 ± 0.07 | 80.69 ± 0.08 | 83.15 ± 0.04 | 87.42 ± 0.17 | 90.46 ± 0.08 | 86.12 |
Rework | 92.22 ± 0.05 | 79.71 ± 0.02 | 81.32 ± 0.03 | 86.64 ± 0.08 | 88.49 ± 0.12 | 84.91 | |
Skip | 91.67 ± 0.07 | 78.22 ± 0.04 | 79.47 ± 0.06 | 82.99 ± 0.07 | 86.98 ± 0.01 | 83.23 | |
Switch | 88.06 ± 0.05 | 54.93 ± 0.17 | 66.89 ± 0.12 | 74.86 ± 0.18 | 75.58 ± 0.15 | 72.70 | |
BPIC12 | All | 87.56 ± 0.01 | 54.23 ± 0.12 | 55.97 ± 0.06 | 61.38 ± 0.05 | 72.11 ± 0.12 | 71.23 |
Rework | 85.84 ± 0.14 | 52,00 ± 0.14 | 52.45 ± 0.01 | 59.30 ± 0.14 | 69.91 ± 0.18 | 68.89 | |
Skip | 83.12 ± 0.08 | 49.87 ± 0.18 | 50.66 ± 0.03 | 56.89 ± 0.05 | 67.44 ± 0.01 | 66.68 | |
Switch | 80.39 ± 0.02 | 40.20 ± 0.03 | 41.71 ± 0.18 | 45.43 ± 0.08 | 57.50 ± 0.02 | 56.74 | |
BPIC17 | All | 91.40 ± 0.05 | 57.51 ± 0.15 | 58.54 ± 0.06 | 64.76 ± 0.12 | 75.63 ± 0.05 | 73.64 |
Rework | 89.98 ± 0.07 | 55.14 ± 0.02 | 56.86 ± 0.07 | 63.94 ± 0.04 | 72.95 ± 0.08 | 70.87 | |
Skip | 89.03 ± 0.01 | 52.67 ± 0.05 | 54.43 ± 0.12 | 60.46 ± 0.02 | 69.75 ± 0.03 | 67.44 | |
Switch | 86.78 ± 0.12 | 42.29 ± 0.03 | 46.95 ± 0.01 | 50.21 ± 0.03 | 58.42 ± 0.01 | 49.26 |