AGCN-T: A Traffic Flow Prediction Model for Spatial-Temporal Network Dynamics

Table 5

Prediction performance of each method on Los-loop.

Time (min)	Evaluation Metrics	Los-loop
		ARIMA	FC-LSTM	T-GCN	ST-GCN	DCRNN	AGCN-T
15	MAE	8.6587	7.5821	6.5607	6.0342	6.3283	5.5518
	MAPE (%)	15.9802	14.9203	15.9824	14.6891	14.0238	15.2011
	RMSE	8.3708	8.9205	8.2458	8.9921	8.5604	8.1386
30	MAE	8.9032	7.9756	6.7432	6.5203	6.0904	5.4398
	MAPE (%)	16.3402	15.9821	15.9903	14.3457	14.0214	15.0389
	RMSE	8.2057	7.9231	8.2459	8.0112	8.2137	8.0911
45	MAE	9.8713	8.9825	6.7943	6.0346	6.9239	5.4472
	MAPE (%)	16.8629	15.7609	15.5914	15.3409	14.5908	15.2411
	RMSE	9.4309	9.4312	8.9403	8.7621	8.9316	8.1582
60	MAE	9.9908	8.4623	7.3568	7.0309	7.4451	5.5311
	MAPE (%)	16.9203	15.3102	15.9213	15.2652	15.5809	15.0778
	RMSE	10.4507	9.0238	9.9522	8.7012	8.9128	8.1084