Scalable Traffic Volume Estimation Using Probe Vehicle Data and Penetration Rate Modeling: Large-Scale Implementation

Accurate, network-wide traffic volume data is essential for effective transportation planning and policy development. While traditional count stations offer reliable measurements, the high cost of their installation and maintenance limits full network coverage. Existing estimation methods often rely on sparse or outdated data, constraining their scalability and accuracy. This study introduces a novel framework that integrates probe vehicle data, stationary counts, and machine learning to estimate traffic volumes via modeled penetration rates. Penetration rates are first computed at locations with observed data and filtered using the interquartile range method to remove outliers. These rates are then used to train an XGBoost model, with Shapley additive explanations employed to assess feature importance and enhance model interpretability. The framework was validated on 675 road segments in Edmonton, Alberta, with 95.11% of observations retained as valid data. A network-wide mean absolute percentage error of 18.19% was achieved, with higher accuracy observed on high-volume roads. These results demonstrate the framework’s scalability and utility for large-scale traffic volume estimation in data-scarce environments, offering transportation agencies a cost-effective alternative to sensor-based monitoring for infrastructure planning and investment prioritization.