Multicomponent Spatial-Temporal Graph Attention Convolution Networks for Traffic Prediction with Spatially Sparse Data

Table 3

Performance comparison of different methods on PeMSD7-sparse. Single Weekly, Single Daily, Single Recent, and Triple Recent are the degraded versions of MSTGACN.

Model	5 min		15 min		30 min
	MAE	RMSE	MAE	RMSE	MAE	RMSE
HA	26.25	40.60	26.25	40.60	26.25	40.60
VAR	20.44	30.40	25.56	37.35	30.60	43.86
STGCN	20.14	29.72	23.43	34.54	26.70	39.03
ST-MetaNet	22.27	32.07	24.45	35.58	26.44	38.91
ASTGCN	21.00	30.55	23.61	34.67	25.98	37.96
DCRNN	19.69	29.40	21.83	33.20	23.48	35.87
Graph WaveNet	19.38	28.83	22.39	32.95	23.43	35.61
Single Weekly (ours)	26.79	40.84	26.46	40.68	26.40	40.50
Single Daily (ours)	26.84	40.68	26.59	40.43	26.68	40.46
Single Recent (ours)	19.23	28.74	21.28	32.29	22.85	34.67
Triple Recent (ours)	19.19	28.57	21.12	32.02	22.64	34.45
MSTGACN (ours)	18.88	28.35	20.65	31.42	21.70	33.33