Multitask Learning with Graph Neural Network for Travel Time Estimation

Table 3

Efficiency of different components.

	Chengdu			Harbin
	MAPE (%)	RMSE (sec)	MAE (sec)	MAPE (%)	RMSE (sec)	MAE (sec)
LSTM for multitask learning	30.14	312.84	241.09	29.47	309.64	232.76
LSTM + GCN + GRU	19.25	286.02	207.98	18.78	262.65	183.91
LSTM + GCN + GRU + GAT	16.48	276.06	188.18	15.16	230.22	167.56
DRTTE (this work)	13.91	252.32	155.71	11.64	242.7	136.29