Ensemble Learning for Short-Term Traffic Prediction Based on Gradient Boosting Machine

Table 2

“5 min ahead” prediction models based on GBDT.

Model	Detectors for prediction	Number of input variables	Spatial factors	MAPE	MAE	Accuracy sorting
Model 1		4	No	0.07652	17.53	11
Model 2		7	Upstream	0.07570	17.3	7
Model 3		7	Downstream	0.07648	17.51	10
Model 4		10	Upstream	0.07480	17.22	3
Model 5		10	Downstream	0.07690	17.61	12
Model 6		10	Upstream and downstream	0.07580	17.32	8
Model 7		13	Upstream and downstream	0.07510	17.3	4
Model 8		13	Upstream and downstream	0.07630	17.48	9
Model 9		16	Upstream and downstream	0.07560	17.4	6
Model 10		13	Upstream	0.07458	16.99	1
Model 11		13	Downstream	0.08583	22.14	15
Model 12		22	Upstream and downstream	0.07738	18.15	13
Model 13		16	Upstream	0.07550	17.17	5
Model 14		16	Downstream	0.08504	21.98	14
Model 15		28	Upstream and downstream	0.07464	17.20	2