Multistep Prediction of Bus Arrival Time with the Recurrent Neural Network

Table 5

Statistics of the prediction results of seven models.

Model	RMSE	MSE	MAE	COS	Trainable parameters	Training time (s)
Pure LSTM	37.7892	1428.029	23.1300	0.9492	20,065	5120
Pure GRU	37.7575	1425.6302	23.0065	0.9494	15,009	5820
LSTM-BP	37.6818	1419.9197	22.8906	0.9498	32,129	6812
GRU-BP	37.8738	1434.4282	22.8810	0.9496	25,089	7886
LSTM-Bi	37.4615	1403.3645	22.5548	0.9507	78,177	9126
LTSM-Stack	34.6279	1199.0963	18.9767	0.9585	525,281	13172
ConvLSTM	34.3812	1182.0734	17.0876	0.9595	117,233	61773

MAE: mean absolute error; RMSE: root mean square error; COS: cosign similarity.