Skip to main content
. 2023 Jul 21;13:11804. doi: 10.1038/s41598-023-38722-5

Table 3.

Comparison of model performance for an average of 5 experimental fills on the traffic flow domain dataset.

Model METR-LA PEMS-BAY
Blocking missing Point missing Blocking missing Point missing
mae mse mape (%) mae mse Mape (%) mae mse Mape (%) mae mse Mape (%)
MEAN 7.48 139.54 12.96 7.56 142.22 13.10 5.46 87.56 8.75 5.42 86.59 8.67
KNN 7.79 124.61 13.49 7.88 129.29 13.65 4.30 49.90 6.90 4.30 49.80 6.88
MF 5.46 109.61 9.46 5.56 113.46 9.62 3.28 50.14 5.26 3.29 51.39 5.27
MICE 4.22 51.07 7.31 4.42 55.07 7.65 2.94 28.28 4.71 3.09 31.43 4.95
VAR 3.11 28.00 5.38 2.69 21.10 4.66 2.09 16.06 3.35 1.30 6.52 2.07
E2GAN 3.00 23.49 5.21 2.98 22.80 7.99 1.97 12.20 3.16 1.77 9.73 2.83
rGAIN 2.90 21.67 5.02 2.83 20.03 4.91 2.18 13.96 3.50 1.88 10.37 3.01
BRITS 2.34 17.00 4.05 2.34 16.46 4.05 1.70 10.50 2.72 1.47 7.94 2.36
SAITS 2.30 16.88 4.00 2.26 16.32 3.94 1.56 14.02 2.50 1.40 7.88 2.30
CSDI 2.23 15.92 3.64 2.20 14.32 3.42 1.50 13.77 2.50 1.22 7.75 1.82
MPGRU 2.57 25.15 4.44 2.44 22.17 4.22 1.59 14.19 2.56 1.11 7.59 1.77
GRIN 2.03 13.26 3.52 1.91 10.41 3.30 1.14 6.60 1.83 0.67 1.55 1.08
ADGCN 2.02 13.22 3.51 1.89 10.31 3.27 1.07 5.23 1.73 0.66 1.52 1.07

The best results are in bold.