In the US, over 38,000 people die in road crashes each year, and 2.35 million are injured or disabled, according to the 2020 Association for Safe International Road Travel (ASIRT) report. At the same time, traffic congestion is keeping Americans stuck on the road, wasting millions of hours and billions of dollars each year. To take steps to improve these statistics, save lives, and reduce the amount of time people must spend in traffic, it is important to understand the root causes. Using statistical techniques and machine learning algorithms, new Mineta Transportation Institute (MTI) research, Analytical Models for Traffic Congestion and Accident Analysis, developed accurate predictive models for traffic congestion and road accidents to better understand the complex causes of these challenging issues.
The research team used four machine learning algorithms to train the predictive models. These are: (1) multivariable logistic regression—a classification algorithm that takes less effort and time to train the model, (2) decision tree—a supervised technique that uses different algorithms to split a node into multiple sub-nodes on all variables and selects the split that provides the best result, (3) Random Forest (RF)—this algorithm eliminates high variance by adding randomness to the model, and (4) Extreme Gradient (XG) boosting—a specific implementation of the Gradient Boosting method. By comparing predictive models trained from four machine learning algorithms, the team found that all methods have similar accuracy using the original imbalanced data, with RF and XG-boost algorithms exhibiting greater accuracy with balanced data.
Results can be used to understand traffic congestion and improve safety on the road. For example, results show:
“This is a highly complex issue,” explains the lead investigator, Dr. Hongrui Liu. “Fortunately, we have a large dataset—4.2 million records of car accidents from 2016 to 2020—to help us train these models and take steps toward improving safety and mobility.”
The predictive models—together with the real time visibility of the environmental conditions offered by advanced information technologies—can help build a smart transportation system.
ABOUT THE MINETA TRANSPORTATION INSTITUTE
At the Mineta Transportation Institute (MTI) at San Jose State University (SJSU) our mission is to increase mobility for all by improving the safety, efficiency, accessibility, and convenience of our nations’ transportation system. Through research, education, workforce development and technology transfer, we help create a connected world. Founded in 1991, MTI is funded through the US Departments of Transportation and Homeland Security, the California Department of Transportation, and public and private grants, including those made available by the Road Repair and Accountability Act of 2017 (SB1). MTI is affiliated with SJSU’s Lucas College and Graduate School of Business.
ABOUT THE AUTHOR
Dr. Hongrui Liu is an MTI Research Associate and assistant professor of Industrial and Systems Engineering at San Jose State University. Rahul Ramachandra Shetty graduated from San Jose State University in Industrial and System Engineering with a specialization in Supply Chain Management. He is currently working as a Supply Chain Business Manager at Lam Research Corporation.
MTI Communications and Operations Manager
|Spatio-Temporal Analysis of Roadside Transportation-Related Air Quality (StarTraq 2021): A Characterization of Bike Trails and Highways in the Fresno/Clovis Area|
|Understanding the Role of Transportation in Human Trafficking in California|
|AI-based Pedestrian Detection and Avoidance at Night using an IR Camera, Radar, and a Video Camera|