Real-Time Prediction of Lane-Based Queue Lengths for Signalized Intersections

Table 3

MAE, MAPE, and RMSE of the predictions of traffic volume proportion in each lane.

Prediction methods	Errors		Lane 1	Lane 2	Lane 3	Average
Kalman filter	MAE (vehs)	All lanes (the proposed method)	2.43	2.37	2.25	2.35
		Single lane (the previous method)	3.36	3.28	2.74	3.13
Single exponential smoothing		Single lane	3.15	2.98	2.46	2.86
Quadratic exponential smoothing		Single lane	2.78	2.77	2.31	2.62
Third-order moving average		Single lane	2.85	2.89	2.35	2.70
Kalman filter	MAPE (%)	All lanes (the proposed method)	17.75	6.52	6.71	10.33
		Single lane (the previous method)	24.96	9.18	8.07	14.07
Single exponential smoothing		Single lane	23.90	8.37	7.13	13.11
Quadratic exponential smoothing		Single lane	20.76	7.67	6.79	11.74
Third-order moving average		Single lane	21.07	8.11	7.07	12.08
Kalman filter	RMSE (vehs)	All lanes (the proposed method)	3.44	3.32	2.70	3.15
		Single lane (the previous method)	4.38	4.34	3.31	4.01
Single exponential smoothing		Single lane	4.61	3.82	3.03	3.82
Quadratic exponential smoothing		Single lane	4.04	3.52	2.76	3.44
Third-order moving average		Single lane	3.82	3.47	2.76	3.35