Attention-Based Gated Recurrent Graph Convolutional Network for Short-Term Traffic Flow Forecasting

Table 2

The prediction performance of different model on the PeMSD4 dataset.

Methods	15 min			45 min			1 h			Training time (s/epoch)
	MAE	RMSE	MAPE (%)	MAE	RMSE	MAPE (%)	MAE	RMSE	MAPE (%)
HA	28.61	42.76	20.00	39.14	57.40	28.4	46.39	67.38	34.60	—
SVR	26.44	41.96	18.50	37.66	57.62	25.7	43.46	66.49	28.90	—
ARIMA	21.8	34.05	14.10	32.91	49.76	21.8	38.97	58.24	26.50	—
MLP	21.47	33.2	14.40	28.67	42.75	20.6	32.77	47.88	25.40	5
GRU	21.63	33.97	15.70	24.11	37.05	17.8	24.56	37.74	18.30	26
T-GCN	22.15	33.88	17.30	28.27	41.99	23.1	31.34	46.47	24.50	30
HGCN	21.1	33.28	14.60	27.81	42.45	19.6	31.45	47.2	23.00	12
ASTGCN	19.9	31.15	14.00	24.26	37.12	17.9	26.8	40.47	20.90	150
AGCRN	19.22	30.08	13.20	21.43	33.29	14.5	22.25	34.42	15.40	110
AGRGCN	20.01	33.03	13.80	21.04	34.65	14.4	21.52	35.28	14.90	51

The bold values indicate the best performance metrics in the performance comparision.