Non-spatial AI modeling to estimate traffic volume measures on local roadways

Accurate estimation of Annual Average Daily Traffic (AADT) on local low-volume roadways remains a critical challenge due to data scarcity and the limitations of traditional approaches. Conventional methods often fail to capture the complex, nonlinear factors influencing traffic volumes, particularly on low-volume, under-monitored roads. This study utilizes a comprehensive dataset from the Smart Location Database (SLD), incorporating geospatial, socioeconomic, and transportation variables to enhance AADT prediction. A Non-Spatial Random Forest (RF) model is developed and benchmarked against traditional Ordinary Least Squares (OLS) models to assess its predictive performance. The RF model significantly outperforms OLS approaches, explaining approximately 72% of AADT variance while capturing complex nonlinear relationships and interaction effects among predictors. Urban accessibility, transit ridership, and walkability emerged as the most influential factors, while workforce and income-related variables showed comparatively weaker impacts. The findings demonstrate that machine learning-based approaches can improve traffic volume estimation in data-limited environments, providing valuable insights for transportation planners seeking to optimize infrastructure investments, enhance mobility, and support evidence-based policy development.