Modelling email traffic workloads with RNN and LSTM models

Table 3 RPEs for all models and all datasets

Prediction error (RPE %)
Category	Technical University of Crete				University of Peloponnese
Category	Previous works		Proposed models		Previous works		Proposed models
Incoming traffic	Probability distribution	RNN	RNN	LSTM	Probability distribution	RNN	RNN	LSTM
Users incoming	21.5	13.9	6.22	5.64	16.7	9.20	3.29	3.33
System incoming	20.8	2.1	1.97	2.04	23.00	7.00	9.96	6.08
Outgoing traffic
Users incoming	14.7	9.4	2.28	1.34	29.60	13.70	4.96	5.22
System incoming	10.00	5.30	2.25	2.22	20.40	4.40	2.83	1.38
Spam traffic
Spam traffic	17.7	17.7	4.71	3.95	25.00	57.10	4.73	4.62

Category	Murdoch University				Liverpool John Moores University
Category	Previous works		Proposed models		Previous works		Proposed models
Incoming traffic	Probability distribution	RNN	RNN	LSTM	Probability distribution	RNN	RNN	LSTM
Users incoming	32.70	14.20	0.37	0.19	8.4	4.2	1.35	1.09
System incoming	9.30	4.20	0.29	0.25
Outgoing traffic					Spam traffic
Users incoming	40.80	25.30	0.97	0.92	36.9	18.7	2.45	2.26
System incoming	22.60	23.30	0.28	0.40