Spatiotemporal DeepWalk Gated Recurrent Neural Network: A Deep Learning Framework for Traffic Learning and Forecasting

Table 2

Performance comparison of ST-DWGRU and other approaches on the datasets PeMSD4, PeMSD8, and PeMS-BAY.

Model	PeMSD4 (15/30/60 min)
	MAE	RMSE	MAPE (%)
HA	2.54	4.96	5.56
ARIMA (2003)	2.51/2.75/3.21	5.72/6.34/7.36	5.32/5.69/6.56
DCRNN (2018)	1.35/1.77/2.26	2.94/4.06/5.28	2.68/3.71/5.10
STGCN (2018)	1.47/1.93/2.55	3.01/4.21/5.65	2.92/3.98/5.39
ASTGCN (2019)	2.12/2.42/2.73	3.96/4.59/5.21	4.16/4.80/5.46
GWN (2019)	1.30/1.70/2.03	2.68/3.82/4.65	2.67/3.73/4.60
LSGCN (2020)	1.45/1.82/2.22	2.93/3.92/4.83	2.90/3.84/4.85
USTGCN (2021)	1.40/1.64/2.03	2.69/3.19/4.25	2.81/3.23/4.32
ST-DWGRU (ours)	1.20/1.48/1.90	2.40/3.12/4.01	2.21/2.75/3.53
Model	PeMSD8 (15/30/60 min)
	MAE	RMSE	MAPE (%)
HA	1.98	4.11	3.94
ARIMA (2003)	1.90/2.12/2.79	4.87/5.24/6.22	5.11/5.21/5.62
DCRNN (2018)	1.17/1.49/1.87	2.59/3.56/4.50	2.32/3.21/4.28
STGCN (2018)	1.19/1.59/2.25	2.62/3.61/4.68	2.34/3.24/4.54
ASTGCN (2019)	1.49/1.67/1.89	3.18/3.69/4.13	3.16/3.59/4.22
LSGCN (2020)	1.16/1.46/1.81	2.45/3.28/4.11	2.24/3.02/3.89
USTGCN (2021)	1.14/1.25/1.70	2.15/2.58/3.27	2.07/2.35/3.22
ST-DWGRU (ours)	1.005/1.25/1.57	2.08/2.70/3.49	1.81/2.24/2.78
Model	PeMS-BAY (15/30/60 min)
	MAE	RMSE	MAPE (%)
HA	2.88	5.59	6.84
ARIMA (2003)	1.62/2.33/3.38	3.30/4.76/6.50	3.5/5.4/8.3
DCRNN (2018)	1.38/1.74/2.07	2.95/3.97/4.74	2.9/3.9/4.9
STGCN (2018)	1.46/2.00/2.67	3.01/4.31/5.73	2.9/4.1/5.4
GWN (2019)	1.30/2.63/1.95	2.74/3.70/4.52	2.7/3.7/4.6
ST-DWGRU (ours)	1.33/1.68/2.08	2.52/3.28/4.17	2.59/3.32/4.14

Bold is the best; underline is the second best.