Data mining for modeling drivers’ behavior in passing a parked vehicle – UROP Spring Symposium 2021

Data mining for modeling drivers’ behavior in passing a parked vehicle

Nathan Tsiang


Pronouns: he,him,his

Research Mentor(s): Brian Lin, Assistant Research Scientist
Research Mentor School/College/Department: University of Michigan Transportation Research Institute, College of Engineering
Presentation Date: Thursday, April 22, 2021
Session: Session 2 (11am – 11:50am)
Breakout Room: Room 18
Presenter: 4

Event Link


This project’s goal has been studying US drivers maneuvering responses to specific driving scenarios which will help in the process of developing automated vehicles. These findings study how drivers will respond as they pass a parked vehicle. This study analyzed data contained within UMTRI’s extensive database containing various measurements of vehicles as they interact with their surroundings. In order to filter the data contained within the UMTRI database, we developed an algorithm using the Structured Query Language. The algorithm uses various parameters to limit the search, such as the dynamic of the ego vehicle (speed, acceleration/deceleration), the interaction between the ego and other vehicles (reaction for brakes, and lateral clearance), and driver’s decision-making. The algorithm discovered around 1,300 instances, of which around 1,050 were correct. MATLAB is being used to perform logistic regression on the studied variables such as speed, curvature, lateral clearance. The variables that have been found to be statistically significant are then analyzed with SVM and decision tree modeling systems. This is to discover whether or not drivers adopted any collision avoidance maneuver when passing parked vehicles, what the deceleration/acceleration profiles of the driver was, and whether the driver braked when passing the target. These results will allow us to conclusively say what kind of responses drivers have to these driving scenarios and provide insights to the development of automated vehicle features.

Authors: Nathan Tsiang, Brian Lin
Research Method: Computer Programming

lsa logoum logo