Traffic volume prediction on low-volume roadways: a Cubist approach

A significant aspect of the U.S. Department of Transportation’s Highway Safety Improvement Program (HSIP) rulemaking is the prerequisite that states must gather and utilize Model Inventory of Roadway Elements (MIRE) for all public paved roads, including low-volume roadways (LVR). States are particularly not equipped with the ability to collect traffic volumes of LVRs due to issues such as budgetary constraints. One alternative is to estimate traffic volumes of LVRs using regression or machine learning (ML) models. The present study accomplishes this by developing a ML framework to estimate traffic volumes on LVRs. By using available traffic counts on low-volume roads in Minnesota, this study applies and validates three different ML models (random forest, support vector regression, and Cubist) to estimate traffic volumes. The models include various traffic and non-traffic (e.g. demographic and socio-economic) variables. Overall, the Cubist model shows better performance compared to support vector regression and random forests. Additionally, the Cubist approach provides rule-based equations for different subsets of data. The findings of this study can be beneficial for transportation communities associated with LVRs.