Deep reinforcement learning for data-efficient weakly supervised business process anomaly detection

Journal of Big Data

Table 6 Balanced accuracy results (%) over all datasets using all the scenarios

Data		Proposed model	Denoising Autoencoder [21]	Variational autoencoder [22]	BINET [5]	Bi-LSTM VAE [23]	Process mining model
Datasets	Scenarios	Proposed model	Denoising Autoencoder [21]	Variational autoencoder [22]	BINET [5]	Bi-LSTM VAE [23]	Process mining model
DS1	All	98.90 ± 0.07	86.62 ± 0.12	89.38 ± 0.18	93.67 ± 0.15	94.95 ± 0.14	92.98
	Rework	95.33 ± 0.07	82.89 ± 0.05	84.51 ± 0.12	90.02 ± 0.07	90.86 ± 0.05	88.23
	Skip	94.41 ± 0.02	81.30 ± 0.01	83.44 ± 0.08	88.45 ± 0.14	89.73 ± 0.07	86.79
	Switch	91.09 ± 0.08	71.47 ± 0.14	72.32 ± 0.06	79.63 ± 0.03	80.24 ± 0.09	78.65
DS2	All	97.02 ± 0.05	84.13 ± 0.15	87.27 ± 0.15	90.54 ± 0.08	93.71 ± 0.15	90.12
	Rework	93.18 ± 0.02	81.67 ± 0.08	82.64 ± 0.09	88.33 ± 0.01	89.87 ± 0.05	86.49
	Skip	92.72 ± 0.04	80.21 ± 0.01	82.04 ± 0.13	85.76 ± 0.00	89.04 ± 0.12	85.87
	Switch	89.61 ± 0.02	57.86 ± 0.05	69.10 ± 0.02	77.15 ± 0.14	77.93 ± 0.09	74.52
BPIC12	All	93.37 ± 0.06	58.03 ± 0.01	60.22 ± 0.05	64.68 ± 0.01	76.00 ± 0.01	74.98
	Rework	88.12 ± 0.08	53.42 ± 0.03	53.58 ± 0.14	61.32 ± 0.12	71.82 ± 0.03	70.63
	Skip	86.98 ± 0.05	52.34 ± 0.09	52.13 ± 0.03	59.93 ± 0.08	69.67 ± 0.00	69.84
	Switch	83.02 ± 0.03	42.97 ± 0.18	42.98 ± 0.17	48.75 ± 0.09	60.40 ± 0.14	58.60
BPIC17	All	95.04 ± 0.04	65.33 ± 0.02	63.46 ± 0.04	67.42 ± 0.19	78.64 ± 0.09	77.42
	Rework	91.84 ± 0.00	61.29 ± 0.17	59.78 ± 0.09	63.33 ± 0.07	74.97 ± 0.14	73.39
	Skip	90.69 ± 0.02	59.81 ± 0.03	57.91 ± 0.03	62.79 ± 0.01	72.91 ± 0.08	71.28
	Switch	87.71 ± 0.12	51.34 ± 0.01	49.11 ± 0.09	52.51 ± 0.15	63.86 ± 0.03	60.74