Application of CNN-LSTM Model for Vehicle Acceleration Prediction Using Car-following Behavior Data

Table 3

The prediction results for four different groups.

Group	Drivers	Model	Index	Experiments					Average
				1	2	3	4	5
1	8 drivers	LSTM	MSE	0.0081	0.0081	0.0082	0.0080	0.0081	0.0081
			MAE	0.0591	0.0592	0.0615	0.0595	0.0606	0.0600
		CNN-LSTM	MSE	0.0076	0.0078	0.0076	0.0076	0.0076	0.0076
			MAE	0.0572	0.0579	0.0574	0.0577	0.0575	0.0575
2	22 drivers	LSTM	MSE	0.0193	0.0218	0.0175	0.0174	0.0170	0.0186
			MAE	0.0891	0.1135	0.0947	0.0907	0.0891	0.0954
		CNN-LSTM	MSE	0.0146	0.0146	0.0148	0.0145	0.0145	0.0146
			MAE	0.0827	0.0822	0.0830	0.0819	0.0828	0.0825
3	30 drivers	LSTM	MSE	0.0173	0.0180	0.0163	0.0173	0.0162	0.0170
			MAE	0.0891	0.0952	0.0869	0.0924	0.0856	0.0898
		CNN-LSTM	MSE	0.0125	0.0127	0.0124	0.0126	0.0128	0.0126
			MAE	0.0746	0.0764	0.0741	0.0754	0.0757	0.0752
4	8 drivers (in random)	LSTM	MSE	0.0219	0.0199	0.0247	0.0203	0.0202	0.0214
			MAE	0.1097	0.1027	0.1191	0.1041	0.1041	0.1079
		CNN-LSTM	MSE	0.0188	0.0188	0.0194	0.0189	0.0193	0.0190
			MAE	0.1007	0.1001	01010	0.1005	0.1008	0.1006

The bold results represent the outcomes of the model with the smallest prediction errors among the four driver groups.