Machine Learning Assessment of Traffic Signals Coupled With a Ranking Algorithm That Relies on Probe Vehicle Data Can Help Agencies Identify Intersections for Signal Retiming.
Machine Learning Methods Were Used to Rank 1,655 Traffic Signals for Signal Retiming in Tennessee.
Date Posted

Using Big Data and Machine Learning To Evaluate and Rank the Performance of Traffic Signals in Tennessee

Summary Information

One basic strategy to improve the performance of signalized intersections is signal retiming. Because examining each intersection manually is very costly and time consuming, many transportation agencies retime the traffic signals every three to five years or rely on citizens’ complaints. In this study, a low-cost database was developed to rank the traffic signals in the state of Tennessee to provide a performance evaluation of each signal and help agencies prioritize projects with traffic signal retiming. Segmented probe vehicle data were obtained from the Regional Integrated Transportation Information System (RITIS) website, and intersections across the state were extracted with Travel Message Channel (TMC) segments. The ranking formula was developed to rank each intersection on a scale of zero to 10 incorporating three performance metrics, namely congestion, planning time index, and bottleneck ranking. 1,655 Tennessee intersections were ranked using available traffic data from September 2021, and stored in a database allowing for easy display, browsing, and querying of the information. A machine learning algorithm was used to divide the signals into six clusters based on the ranking factors. Results from this classification were compared with the level of service (LOS) letter grade of 55 intersections in the Cities of Murfreesboro and Franklin provided by local transportation agencies.

Lessons Learned

  • Monitor traffic signal performance in real-time. In this project, signal ranking results were calculated utilizing historic traffic data from September 2021. To more accurately represent the current roadway network conditions, providing real-time score using data pulled from the RITIS website periodically or implementing an Automated Traffic Signal Performance Measures (ATSPM) system is recommended for the long run.
  • Include safety in the ranked choice factors. Safety was not included in this study due to availability of data but should be implemented into signal assessment technologies.
  • Ensure side roads are included in the TMC segments. In this project, the TMC segments used to extract intersections are not available for all roads, especially side roads, which limited the number of intersections in the database. Including side roads in the TMC segments would expand the database to more traffic signals.
  • Consider machine learning for evaluating traffic signal performance. Machine learning can fine-tune the weighting of the different ranked choice factors in the calculation of signal priority and refine the algorithm.
  • Conduct more outreach with local transportation agencies. Local transportation agencies should be made aware of technologies to assess signals, and dialogue channels should be established for officials to provide feedback on the ranking system.
System Engineering Elements

Keywords Taxonomy: