A data-driven approach to characterize the impact of connected and autonomous vehicles on traffic flow

Parsa, Amir Bahador; Shabanpour, Ramin; Mohammadian, Abolfazl (Kouros); Auld, Joshua; Stephens, Thomas · 2021 · Crossref

DOI: 10.1080/19427867.2020.1776956

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of characterizing the impact of connected and autonomous vehicles (CAVs) on traffic flow, a problem complicated by the lack of large-scale historical data. The authors aim to develop a data-driven model that relates changes in network traffic flows to traffic network and built-environment characteristics. By establishing this relationship, the model can predict traffic flow changes in geographical contexts where CAV-based network simulators are unavailable. The research focuses on the Chicago metropolitan area, using simulation outputs to serve as a proxy for historical data. The methodology employs the POLARIS agent-based modeling platform to simulate traffic under two extreme scenarios: 0% CAV penetration (base) and 100% CAV penetration. The dependent variable is the change in average daily traffic (ADT) across 22,465 network links. To predict these changes, the authors engineered features from multiple data sources, including link properties (road type, lanes, slope), network metrics (connectivity index, distance to central business district, road and intersection density), demographic data (population, vehicle ownership, job density), transportation indicators (trip equilibrium index), and land-use patterns within 150-meter buffers of each link. Three machine learning models—K-Nearest Neighbors (KNN), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)—were trained using a 70/30 train-test split with 5-fold cross-validation. SHapley Additive exPlanations (SHAP) were used to interpret feature importance. The results indicate that XGBoost outperformed the other models, achieving an accuracy of 89.7%, compared to 87.1% for RF and 83.5% for KNN. Analysis of the simulation outputs revealed that ADT increases significantly under the CAV scenario, particularly on freeways and expressways connecting downtown to suburban areas. SHAP analysis identified link properties, specifically road type and number of lanes, as the most influential predictors. Freeways and expressways with more lanes experienced the highest increases in ADT. Gross population density was the next most significant feature, showing an inverse relationship with ADT changes; roads in denser areas saw smaller increases. Distance to the CBD also negatively correlated with ADT changes, as roads closer to the CBD already had high traffic volumes. Other significant factors included intersection density, job opportunities near roadways, and household vehicle ownership. The significance of this work lies in providing a validated, data-driven framework to quantify CAV impacts on traffic flow without requiring complex agent-based simulations for every new region. The study demonstrates that machine learning can effectively map network and environmental characteristics to traffic flow changes. This approach offers a scalable tool for transportation planners to assess CAV impacts in areas lacking specific simulation infrastructure, facilitating broader application of CAV impact analysis across different geographical contexts and penetration levels.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• traffic density