Research Article

Implications of Spatiotemporal Data Aggregation on Short-Term Traffic Prediction Using Machine Learning Algorithms

Table 8

Comparison of feature selection methods.

ModelFeaturesAccuracyRMSEMAE
CorrelationRandomRFECorrelationRandomRFECorrelationRandomRFE

ANN100.8330.8000.85570.180.758.349.859.143.4
200.8450.8220.88066.169.447.846.553.635.8
400.8610.8430.89160.263.845.542.647.633.3
600.8570.8640.90162.855.641.143.841.630.1
800.8610.8570.90661.460.639.842.943.528.7
1000.8620.8920.90660.445.339.042.332.828.3
1200.8610.8830.90761.149.540.042.735.928.8
1400.8560.8860.90763.548.240.044.135.028.4
1600.8740.8750.90555.654.040.039.138.429.0
1800.8680.8760.90257.252.143.940.238.031.0

RF100.8320.7970.87569.281.152.450.361.138.5
200.8420.8360.88665.962.948.047.848.934.7
400.8470.8580.89464.055.543.946.042.132.2
600.8520.8560.89263.758.744.645.143.732.8
800.8510.8650.89364.057.244.545.241.132.6
1000.8520.8680.89363.854.244.245.040.432.5
1200.8510.8600.89364.057.644.245.242.332.5
1400.8520.8680.89363.854.044.045.040.432.4
1600.8510.8670.89464.053.943.945.140.232.4
1800.8520.8720.89363.952.544.144.938.732.4

SVR100.8250.7220.85072.6114.159.353.084.445.2
200.8220.7510.88073.5100.847.253.473.136.5
400.8340.7960.87469.481.650.050.160.437.7
600.8340.8160.87870.276.350.050.055.237.2
800.8320.8380.87971.566.049.950.548.736.9
1000.8320.8420.87972.063.850.350.547.537.1
1200.8290.8280.87772.770.050.851.251.837.5
1400.8300.8320.87672.669.351.351.150.737.8
1600.8290.8390.87572.765.951.751.148.338.0
1800.8300.8350.87672.666.951.351.149.537.7

Traffic volume predictions under nonrecurring conditions with input data; given a constrained number of features, in most cases, the RFE method achieves better performance compared to random and correlation-based feature selection.