Data mining for modeling drivers’ responses to cutting-in and out of traffic – UROP Spring Symposium 2021

Data mining for modeling drivers’ responses to cutting-in and out of traffic

Jason Hu

Jason Hu

Pronouns: he/him

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: 3

Event Link


Human drivers often perform and react to high speed maneuvers unpredictably, cutting in and out of traffic on highways with varying rates, acceleration, reaction time, and more. By analyzing naturalistic driving data from the University of Michigan Transportation Research Institute database, it is possible to determine some relevant distributions and tendencies of these variables. Data was queried from a large database with Structured Query Language, and then processed in MATLAB to find instances of vehicles cutting in and out. After the cutting events were manually verified, culminating in 857 cut in events and 866 cut out events, the variables associated with the events, including range, range rate, lateral rate, and average acceleration were analyzed using methods from statistics and machine learning, including logistic regression, support vector machine, and decision trees. Correlations between key variables were found, as well as models for predicting the impact of cutting events on traffic. More specifically, range and range rate were found to be significant variables in determining the braking behavior of a vehicle in response to a cut in, with decision trees and variants being used to predict the behavior. These results have implications on future design of autonomous vehicles, such as the safe integration with human drivers and minimizing risk during partially unpredictable events.

Authors: Jason Hu, Jim Vega, Brian Lin
Research Method: Computer Programming

lsa logoum logo