Big Data Methodologies for Simplifying Traffic Safety Analyses

Project Description Due to the low rate of crashes/near-crash events, most drivers and majority of trips are event-free. This high imbalance is challenging to researchers as traditional statistical methods and data mining tools are not successful at dealing with rare event data. Until now, there has been little research done on this matter, making SHRP2 NDS data a great resource for this study. SHRP2 NDS data as it is not only the largest NDS to date, this collection contains 3,400 drivers, 35 million miles of continuous driving as well as 1,500 crashes and thousands of near-crashes. This study is focused on identifying kinematic variables and optimal threshold values that predict high-risk drivers using rare event modeling as well as a variety of other statistical and data mining models. It has been shown that driver's higher elevated gravitational-force rates are highly correlated with their crash/near-crash rate, making it a favorable characteristic to asses risk on a driver level. Analysis will be conducted on a trip level to eliminate environmental confounding variables such as; driving time, average speed, road conditions, etc.. Data Request Scope The scope of the data requested is primarily kinematic information of high g-force, trip summary, event, driver demographic, and basic vehicle information. Data Specifications Driver Demographics Questionnaire Sleep Habits Questionnaire Driving Knowledge Survey Clock Drawing Score Barkley's ADHD Screening Test Sensation Seeking Scale Survey Driver Behavior Questionnaire Event Details Table Trip Summary Table Vehicle Details Table High G-force Event Table

搜集汇总

背景与挑战

背景概述

该数据集源自SHRP2自然驾驶研究（NDS），涵盖3400名司机、3500万英里驾驶记录以及1500次碰撞和数千次接近碰撞。其特点是数据高度不平衡（罕见事件），旨在通过罕见事件建模等方法识别预测高风险驾驶员的运动学变量和阈值。数据包括高g力事件、行程摘要、驾驶员人口统计和车辆信息等。