An Ensemble Learning Model of Crash Severity within Georgia I-285@I-20 East Interchange project area

The primary goals of Georgia DOT’s I-285@I-20 East Interchange project are to improve traffic flow and safety by expanding and upgrading the existing multi-level interchange and nearly ten miles of these two interstates that were initially constructed more than a half-century ago to long-outdated design standards.  To optimize the safety and quantify the expected reductions in crashes, a machine learning based model (XGBoost model) investigates and identifies significant contributing factors that determine the crash severity near the project area. Crash data from Georgia Electronic Accident Reporting System (GEARS) database from 2018 to 2022 have been collected and are used in this study. Crash injury severities are classified into five categories: no injury (property damage only), injury class 3 (possible injury), injury class 2 (evident injury), injury class 1 (disabling injury), and fatality. The outputs of the developed XGBoost model are evaluated and compared to the traditional Multinomial logit (MNL) modeling results. The developed model and analysis results will provide insights on developing effective measures to reduce crash severities and improve traffic safety.