Estimating Traffic Volume of Nonstate Roadways with Support Vector Regression

Annual average daily traffic (AADT) is a critical input to many key components of transportation activities. Accurate AADT data are vital to the calibration and validation of travel demand models, roadway improvement funding allocations and safety performance evaluations. Non-state roads constitute a large percentage (usually 60 to 70%) of the total mileage of a state’s roadway network. Traffic volumes on these roads are generally low, and the vehicle miles traveled (VMT) on these roads is much less compared with that on interstate or arterial roads. Thus, regularly conducting traffic counts is not economically feasible for non-state roadways. This study develops an AADT estimation methodology using Support Vector Regression (SVR). By using available traffic counts at block level on non-state roadways and four variables - namely population, job, distance to intersection, and distance to major state highways at block level - a method to estimate roadway AADT for eight parishes was developed. The current study contributes to obtain better AADT estimates than the conventional parametric statistical methods. With the estimated AADT, the local government agencies can make better decisions on funding allocations for safety improvement projects and pavement maintenance actions. The estimated disaggregate level AADT on non-state local roads can also improve statewide travel demand forecasting models.