Measuring the Last-Mile: Leveraging Synthetic Data to Evaluate the Effects of Urban Freight Interventions

Term Start:

June 1, 2025

Term End:

August 31, 2026

Budget:

$160,000

Keywords:

Machine Learning, Synthetic Freight Data, Truck GPS Modeling

Thrust Area(s):

Data Modeling and Analytic Tools

University Lead:

City College of New York

Researcher(s):

Alison Conway

In most urban areas, there is currently a misalignment between difficult-to-measure freight vehicle demands and infrastructure capacities (e.g., parking and loading zones) to accommodate these demands. This misalignment typically results in high costs for industry (e.g., wasted time, wasted fuel, parking fines) as well as in congestion, traffic conflicts, and other externalities for the surrounding community. To better manage this supply-demand mismatch, many cities in the US (e.g., New York, Seattle, Philadelphia, Miami, Los Angeles, Washington) and around the world are exploring policy interventions (e.g., active curb management approaches, local access restrictions, pricing) or city logistics strategies (e.g., off-hour deliveries, micro-hubs, parcel lockers, consolidation centers) to induce freight travel behavior changes – such as mode or vehicle type shifts, temporal shifts, or collaboration between carriers – or to better align freight transportation supply and demand. However, local governments often lack access to the granular data needed both to understand baseline conditions and to simulate the effects of such policy and infrastructure interventions or new city logistics approaches at a system scale. While granular location-based data can be used to derive detailed information about truck travel activity, carriers are often hesitant to share such data with public agencies due to data privacy concerns.

Building on results from our previous project reviewing potential methods to synthesize local freight data useful for decision-making, the goal of this project is to develop, implement, and evaluate a machine-learning based framework to synthesize an anonymized truck GPS dataset with suitable granularity for evaluating urban freight policy interventions. We will develop a simulation approach that leverages appropriate machine learning techniques for encoding/decoding real truck GPS data, enriched by other publicly-available information, to simulate detailed path segments between traffic analysis zones (TAZs). We will implement additional machine learning techniques to assign these segments to sequences, constrained by other aggregate data sources, to produce realistic, high-resolution synthetic truck paths. This synthesized data can be used to characterize individual truck activity as well as to evaluate aggregate activities within a TAZ.

Using the synthesized data, we will then develop three case studies to evaluate the utility of the synthetic data produced to measure and characterize local truck travel behavior, and to quantify traffic, parking, air pollution, or other effects for before-after implementation scenarios for each case study.

Project scope

Scroll to Top