基 于 选 择 集 成 的 山 区 高 速 事 故 预 测 模 型

To improve the prediction accuracy and reduce the robustness of the traffic accident prediction model，this paper uses the Stacking integration strategy to construct an integrated traffic accident prediction model. Firstly，single traffic accident prediction models based on eight machine learning models，such as Decision Tree and Extra Tree，were constructed and the MIC test was used to measure the similarity of each traffic prediction model with the graph coloring method，and the models with low similarity and high diversity were selected to participate in the integration. Secondly，Box-Cox transformations were applied to the results of the single accident prediction models and different weights were assigned to each single model separately using feature weighting method. Finally，models such as BP neural network and Logistic regression were selected as meta-learners for Stacking integration. The results of the study show that the prediction accuracy of the integrated model with BP neural network selected for the meta-learner is higher than other integrated models，and the MAE and RMSE of the integrated model have been respectively reduced by 24% and 14% and the R² has been improved by 6% compared to the single accident prediction model with the highest prediction accuracy.