Examining Complex Impacts of E-shopping and Built Environment Factors on Shopping VMT
December 28, 2024
First author of a peer-reviewed study published in Transportation Research Part D, analyzing the impacts of online shopping on shopping-related vehicle miles traveled (VMT) using machine learning.
Publication Details for Transportation Research Paper:
Published In: Transportation Research Part D: Transport and Environment
Volume: 139, February 2025, Article 104567
Authors: Manreet Sohi, Patrick Loa, Basar Ozbilen, Xiatian Iogansen, Yongsung Lee, Giovanni Circella
Read the full paper on Transportation Research Part D. If you’d like a copy and do not have access, feel free to contact me directly."
Motivation & Overview
This research project was fueled by my personal passion. I had visited Punjab the winter before starting this project—my first visit in a while—and I was shocked by the severe pollution. New Delhi has some of the worst pollution in the world, a problem that extends to Punjab, especially in larger cities like Ludhiana. The air felt suffocating even in winter. While Punjab typically has hot, humid summers, I was particularly struck by how oppressive the traffic-related smog had become. This experience sparked my curiosity about transportation systems, leading me to join ITS as a student research assistant. Initially, my role involved basic tasks like inputting survey data, but after expressing my interest in transportation research and highlighting my computer science background, I was kindly given the opportunity to lead my second research paper. It turned into an incredible experience. I began the paper in May 2024, received an National Center for Sustainable Transportation (NCST) fellowship Summer 2024 to fund my work, and recently celebrated its publication. I'm incredibly grateful for this journey.
As the lead author on my second published research paper, I took on significantly more responsibility this time. The paper was more substantial than my previous one, as I was responsible for leading statistical analysis in R, developing the LightGBM machine learning model in python—spending about 2 months on model development, fine-tuning hyperparameters, and variable optimization, while finishing up my final quarter of university. Once we were satisfied with the model's performance, I led the writing process such created the first draft and making edits during the submission process, creating SHAP plots and calculating the average treatment effects to further expand on the policy recommendations. I'm grateful for this collaborative experience with established postdoctoral researchers at ITS (Institute of Transportation Studies). Their combined expertise in transportation studies, machine learning, and environmental assessment strengthened our methodology and findings.
Key Findings
Machine Learning Model Performance: The LightGBM model demonstrated superior predictive accuracy compared to traditional regression methods, achieving a 15% lower Mean Absolute Error (MAE) and capturing complex non-linear relationships between online shopping behavior and VMT that linear regression failed to identify.
Online Shopping Effects: revealed contrasting impacts - Food-related online shopping (groceries, restaurant orders) showed a substitution effect. Non-food online shopping (clothing, electronics) demonstrated a complementary effect.
Built Environment Influence: Higher density neighborhoods (>10,000 people/sq mile) and frequent transit service (>20 trips/hour) were associated with 25% lower shopping VMT compared to low-density areas, highlighting the importance of urban form in sustainable transportation patterns.
Policy Recommendations
Enhance public transit infrastructure and frequency while promoting mixed-use, high-density development
Establish centralized delivery hubs near transit stops to reduce individual delivery trips
Design walkable neighborhoods with easy access to retail and active transportation options
Incentivize electric and low-emission vehicles for both personal and delivery use
These policy recommendations aim to balance the convenience of online shopping with environmental sustainability goals by reducing overall VMT and associated carbon emissions.
Skills Applied:
Machine Learning (LightGBM): Developed LightGBM model in Python with Optuna optimization
Data Analysis and Visualization (SHAP): Created SHAP visualizations and calculated ATEs to analyze variable relationships
Sustainable Transportation Research: Led research methodology and technical writing, publishing in Transportation Research Part D
Policy and Urban Planning Insights: Created data-driven policy recommendations for sustainable transportation and VMT reduction
End result: This research is a pivotal step toward understanding and addressing the environmental impacts of our increasingly digital economy, suggesting practical policies to decrease greenhouse gas emissions from transportation.