Deploying an Artificial Intelligence-based Decision Support System (AI-DSS) with Variable Speed Limits Led to a 14 Percent Reduction in Crashes on I-24 Corridor in Tennessee.
A Before-and-After Evaluation Compared Safety, Mobility, Efficiency, and User Satisfaction Metrics Based on 1.5-Year Deployment Data.
Tennessee, United States
ATCMTD Final Report for the TDOT Artificial Intelligence-Based Decision Support System (DSS)
Summary Information
The Interstate 24 (I-24) corridor between Nashville and Murfreesboro, Tennessee, has experienced persistent congestion and safety challenges, creating a need for operational strategies that could improve corridor performance. To address these challenges, the Tennessee Department of Transportation (TDOT) and its partner agencies implemented intelligent transportation systems (ITS) and integrated corridor management (ICM) strategies to enhance safety and efficiency along this corridor. Funded under the Advanced Transportation and Congestion Management Technologies Deployment Initiative (ATCMTD) grant program in 2021, TDOT deployed an Artificial Intelligence-Based Decision Support System (AI-DSS) to support real-time traffic management. Compared with microsimulation-based decision support systems (DSS), the AI-DSS was designed to be more economical, efficient, and transferable. The system utilized various transportation management technologies including variable Speed Limit (VSL) signs, Lane Control Signs (LCS), Dynamic Message Signs (DMS), video detection for emergency pull-offs, upgraded and connected traffic signals, new closed-circuit television (CCTV) cameras, and Radar Detection Systems (RDS). The AI-DSS gathers traffic and incident data from field monitoring devices and TDOT’s SmartWay Central Software, analyzes corridor conditions in real time, and provides recommended actions to the Transportation Management Center (TMC). These actions included posting VSLs, sharing traveler information, supporting lane-control operations, and recommending traffic signal timing changes. The deployment extended along I-24 between the I-440 and I-840 interchanges, where TDOT constructed 67 overhead gantries placed approximately every half mile in each direction.
METHODOLOGY
To assess the impact of AI-DSS system, the project evaluated safety, mobility, TMC efficiency, and user satisfaction. Specifically, this project examined the before-and-after impacts of the AI-DSS-controlled VSL. Key indicators included frequency and severity of crashes, traffic delays from recurring and non-recurring congestion, incident management, benefit-cost-ratio (BCA), and user and operator feedback. Evaluation data were collected from 2.5-year pre-deployment and 1.5 years post-deployment and drawn from three primary sources: TDOT SmartWay database, AASHTOWare Safety database, and Regional Integrated Transportation Information System (RITIS) Probe Data database. The evaluation mainly used a before-and-after comparison design across multiple deployment phases:
- I-24 ICM Go-Live (06/20/2023-09/17/2023)
- Initial AI-DSS Go-Live (09/18/2023-02/28/2024)
- Final AI-DSS Go-Live (03/09/2024-12/31/2024)
Each phase was compared against baseline averages from equivalent calendar periods in prior years. Safety and incident detection measures were annualized, while mobility and incident clearance metrics were averaged for comparison.
FINDINGS
- The activation of VSL led to fewer crashes. When the VSL system was active, the crash rate fell 14 percent (from 18.4 to 15.8 crashes per month) and the secondary crash rate fell 50 percent (from 7.2 to 3.6 crashes per month). Primary crashes decreased by seven percent after deployment (from an annualized 1,811 to 1,679), rear-end collisions decreased by 10 percent (from 770 to 697 per year), and fatal collisions decreased by 11 percent (from 9 to 8 per year), with total crash costs declining 10 percent from approximately $290.7 million to $262.6 million annually.
- The evaluation of AI-DSS yielded a benefit-cost-ratio (BCR) of 4.98, which is greater than that of typical roadway widening project (generally 1.0-2.0). Project benefits were around $30.4 million per year in 2023 dollars and indicated a break-even point of approximately 2.5 years, primarily driven by a reduction in user delay costs and crash costs. Over a 15-year lifespan with a 3.1 percent discount rate, total net benefits were estimated at $370,981,135.55.
- The TMC operation achieved reduced incident clearance times and received positive TMC technician feedback. VSL responded on average nine minutes before a crash was reported to the TMC, representing a 90 percent reduction in warning response time, with correct warnings issued 88 percent of the time and a false warning rate of only 1.6 percent.
- Incident clearance time decreased by 20 percent (from 333 to 267 minutes) and annual incident detections increased by 16 percent (from 8,200 to 9,494) following deployment. Faster incident detection also improved technicians’ efficiency.
- Traffic volume increased eight percent from baseline to post-deployment with negligible change in average travel times, indicating improved corridor throughput.
