详情

原文传递 Lane Change Decision Analysis on Freeways by LightGBM and SHAP

论文题名：	Lane Change Decision Analysis on Freeways by LightGBM and SHAP
关键词：	Freeways;Lane Change;Decision Analysis;Light GBM;Light SHAP
摘要：	Vehicle lane change maneuvering could have a substantial influence on traffic flow and safety as a result of their interfering influence in the surrounding lane. The interference effect of lane changing could be more pronounced when vehicles maneuvers horizontally and change the white road marking. Most freeway crashes occur in the vicinity of interchange diverging and merging areas, especially in speed-change lanes. Different vehicles maneuvers in different lanes during a lane change. This alteration in maneuvering is due to the influence of surrounding vehicles which impose on the lane change vehicle. This study is exploring the factors attribution in making lane change prediction through Shapley Additive Explanation. So, in order to predict the maneuvering in terms of lane changes on the urban arterial, accurate machine learning modelling is required. 　　For developing machine learning models, real data are used to set up explanatory variables and to the evaluation of models. This study will help traffic engineers and manager to a better understanding of the how various lane change maneuvering takes place and the impact of lane change on traffic operation and safety. The results showed that the trained logistic regression model achieved an accuracy of 70%, while the accuracy of Light GBM classifier with tunned hyper-parameters was 73.4%in predicting the lane changing instant and the surrounding traffic characteristics on lane change vehicles. Area under the ROC was pretty satisfying. 　　From the SHAP value analysis, it was found that the vehicle dimensions (length, Width) and velocity shows a significant impact on lane changing decision, and also lane changing with insufficient space headway gap makes the collision risk higher and driver engage in lane changing process. With larger space headway lane change urgency reduces. The same behavior has been found with the following gap in current lane and gap acceptance in the target lane. In the case of preceding and following vehicle speed in the current and target lane have also a great contribution in predicting lane change. 　　The lane changing behavior shows that the more frequent is the lane change the chances of vehicle conflict is higher. It is evident that higher vehicle speed contributes to more crash risk with the following vehicle in current and target lane, while lower speed has a safer traffic condition. In the case of flow rate, more frequent lane change has a negative impact on traffic flow.
作者：	Muhammad Owais Latif(欧瓦)
专业：	交通运输工程;交通信息工程及控制
导师：	Chen Wang
授予学位：	硕士
授予学位单位：	东南大学
学位年度：	2021