Key points are not available for this paper at this time.
To improve the prediction accuracy of traffic flow, a travel time prediction model based on gradient boosting decision tree (GBDT) is proposed. In order to test the applicability of GBDT, models with different prediction horizons (5 min ahead, 10 min ahead, and 15 min ahead) are established. The 11 variables are viewed as candidates in this paper. Different from other machine learning algorithms as black boxes, GBDT can provide interpretable results through variable importance. In the proposed model, the variable importance shows that for different prediction horizons, the most important influence variable is uniform, which is travel time in the current period. Traffic conditions in the current period have the greatest influence on the predicted travel time. Compared with the back propagation neural network model and the support vector machine model, the proposed GBDT model can produce more accurate prediction results, especially in multi-step prediction, indicating that GBDT is a promising method in travel time prediction.
Cheng et al. (Fri,) studied this question.