中文摘要：

采用基于贝叶斯方法的决策树算法，利用上海市中心城区1536个交通事件持续时间数据，建立交通事件持续时间的预测模型。结果表明，事件类型是决策树中的第一层测试属性，不同类型事件的特性属性在决策树中的位置并不相同。并用384个交通事件数据对模型的预测精度进行检验。检验结果表明，抛锚事件持续时间预测误差小于10min的正确率为79％，而交通事故持续时间预测误差小于20min的正确率为65％。基于贝叶斯推理的决策树算法比仅基于贝叶斯或仅基于决策树算法的分类精度更高，鲁棒性更强。

英文摘要：

The paper presents a prediction method of traffic incident duration of expressway, grounded on the Bayesian method-based decision tree classification algorithm and 1536 in cident data of Shanghai central city expressway. And 384 incident data were adopted to test the prediction accuracy of this model. The results show that the incident type is the first layer of the decision tree and different incident has different test attributes. The prediction accuracy of anchor duration is 79 % with an error of 10 minutes while that of accident duration is 65 % with an error of 20 minutes. So the Bayesian method-based decision tree algorithm is more accurate and stabilized than the method based on Bayesian method or decision tree respectively.