Extracting Overtaking Segments by Unsupervised Clustering and Predicting Nonmotorized Vehicle’s Trajectory

Table 4

Comparisons on clustering method, model structure, and input version.

Clustering method	Model structure	Input variables	Root-mean-square error	Mean absolute error	Mean absolute percentage error
HDP topic model	Conv-LSTM-LSTM	Lite	0.091	0.050	1.826
	Conv-LSTM-LSTM	Full	0.517	0.296	12.331
	Conv-LSTM-LSTM-LSTM	Full	0.592	0.340	17.951
	Conv-GRU-GRU	Lite	0.145	0.073	3.329
	Conv-GRU-GRU	Full	0.555	0.298	15.602
	Conv-GRU-GRU-GRU	Full	0.725	0.460	25.456
HDPHMM	Conv-LSTM-LSTM	Lite	0.073	0.041	1.313
	Conv-LSTM-LSTM	Full	0.300	0.104	3.055
	Conv-LSTM-LSTM-LSTM	Full	0.391	0.198	7.679
	Conv-GRU-GRU	Lite	0.099	0.050	1.790
	Conv-GRU-GRU	Full	0.603	0.315	12.546
	Conv-GRU-GRU-GRU	Full	0.412	0.217	6.563