Research Article
Dynamic Path Flow Estimation Using Automatic Vehicle Identification and Probe Vehicle Trajectory Data: A 3D Convolutional Neural Network Model
Table 2
Test of bootstrapping algorithm for different percentages of noisy labels.
| | | Percentage | 10% | 30% | 50% | 70% | 90% |
| | Part 1 (with pretraining) | RMSE | (3.63/3.55) | (4.29/3.55) | (5.37/3.56) | (6.57/3.58) | (8.08/4.45) | | RMSE% | (45.56/44.50) | (53.84/44.50) | (67.35/44.60) | (82.37/44.88) | (101.35/55.83) | | MAE | (1.89/1.89) | (2.05/1.89) | (2.36/1.89) | (2.75/1.91) | (3.27/2.06) | | MAE% | (23.73/23.72) | (25.67/23.72) | (29.64/23.75) | (34.50/23.94) | (41.07/25.89) |
| | Part 2 (without pretraining) | RMSE | (3.71/3.74) | (4.36/3.8) | (5.43/3.99) | (6.57/4.48) | (8.12/7.79) | | RMSE% | (46.60/46.87) | (54.72/47.69) | (68.18/50.09) | (82.42/56.18) | (101.89/97.68) | | MAE | (1.90/1.93) | (2.05/1.94) | (2.35/1.97) | (2.72/2.04) | (3.28/2.68) | | MAE% | (23.86/24.26) | (25.7/24.34) | (29.52/24.8) | (34.14/25.65) | (41.14/33.66) |
|
|
