Quality assurance for network-level traffic signal performance measures derived from connected-vehicle data

Automated traffic signal performance measures (ATSPMs) have become critical for traffic operation and management by providing a series of performance measures that quantify the traffic conditions at signalized intersections. However, collecting these performance measures requires upgrading existing traffic detection systems, which is time-consuming and costly. Therefore, connected-vehicle data has been used as a substitute data source for obtaining performance measures, saving money and time. However, a limited number of studies have investigated controlling the data accuracy of performance measures obtained from connected-vehicle data. Considering this issue, we propose a method to determine the sample size indicator and associated threshold for control delay, arrival on green (AoG), and split failure performance measures calculated from connected-vehicle GPS data, specifically Wejo data. A comparison of the trajectory-based performance measures and traditional sensor-based performance measures shows that the trajectory-based delay and AoG for through movements require at least 16 vehicles/hour to reach the maximum correlation coefficient. The trajectory-based delay on left-tun movements needs at least 6 vehicles/hour/lane to reach the maximum correlation coefficient. The two types of split failure data show a weak correlation because of excess zero split failure, which is challenging for determining the sample size threshold. In addition, logistic regression is developed to investigate and quantify the impacts of various factors, including temporal factors, socio-economic factors, road characteristics, and traffic conditions, on the data availability and sufficient sample size.