论文部分内容阅读
为了进一步提高决策树模型的交通事件检测性能,且避免噪音和过拟合现象,提出了基于随机森林的交通事件检测方法.从分类强度和相关性2个角度进行分析,并构建了3组实验:与不同数目决策树的对比、与不同决策树的对比及与神经网络的对比.实验数据采用实测的高速公路交通参数数据库(I-880数据库);实验的评价指标采用检测率、误警率、平均检测时间、分类率和ROC曲线下的面积.实验结果表明,基于随机森林的交通事件检测模型可以提高检测率、减少检测时间、提高分类正确率,和多层前馈神经网络相比具有很好的竞争力.
In order to further improve the traffic incident detection performance of the decision tree model and avoid the noise and overfitting phenomenon, a traffic incident detection method based on random forest is proposed.From the perspectives of classification strength and correlation, two sets of experiments are constructed and three sets of experiments : Comparison with different number of decision trees, comparison with different decision trees and comparison with neural networks Experimental data using the measured highway traffic parameter database (I-880 database); the evaluation index of the experiment adopts the detection rate, false alarm rate , Average detection time, classification rate and area under the ROC curve.Experimental results show that the traffic incident detection model based on random forest can improve the detection rate, reduce the detection time and improve the classification accuracy, compared with multilayer feedforward neural network Very competitive.