Description

Vehicle miles traveled (VMT) is a fundamental metric for assessing mobility trends and infrastructure needs. This study examines the spatial-temporal dynamics of daily VMT (DVMT) and per capita DVMT across California counties from 2019 to 2023, covering the pre-, mid-, and post-pandemic periods via GIS mapping and k-means clustering. To identify determinants of per capita DVMT, we compared traditional linear regression approaches (OLS, Ridge, LASSO, Elastic Net) with ensemble tree-based models. Specific results include: the ensemble models estimated using 2019–2022 data delivered substantially higher accuracy, achieving R² values exceeding 0.98; meanwhile, out-of-sample performance on 2023 data remained robust (R² ≈ 0.82 for Random Forest; R² ≈ 0.91 for Gradient Boosting), indicating strong model generalizability. Feature importance analysis identifies housing density, population density, and public transit mode share as the primary drivers of per capita DVMT. These findings underscore the utility of spatial analysis and advanced nonlinear modeling for regional transportation planning.

Publication Date

4-10-2026

Publication Type

Report

Topic

Planning and Policy, Sustainable Transportation and Land Use, Transportation Engineering, Transportation Technology

Digital Object Identifier

10.31979/mti.2026.2475

MTI Project

2475

Keywords

Vehicle miles traveled, K-means clustering, Regression analysis, GIS

Disciplines

Transportation

Download

Additional files available below

Share

COinS