Learning Transportation Insecurity from Location Intelligence Data

Nie, Yu (Marco); Chen, Ying; Murphy, Alexandra K.; Dai, Tianxing; Yan, Xiaoyu; Kivestu, Peeter · 2026 · ROSA P / University of Michigan. Center for Connected and Automated Transportation

archive: archived pipeline: cataloged verified

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

Summary

This paper addresses the challenge of measuring transportation insecurity (TI)—a condition where individuals lack the resources to move safely and timely—at a scalable level. While the Transportation Security Index (TSI) provides a validated measure of TI, its reliance on stand-alone surveys limits temporal coverage, geographic comparability, and scalability. The authors propose a transfer learning framework that leverages large-scale location intelligence data (mobile phone trajectories) to infer TI status, decoupling measurement from repeated survey administration. The core hypothesis is that mode use patterns, which are strongly associated with TI in survey data, can be extracted from passive mobility data to predict insecurity across broader populations. The methodology employs a semi-supervised machine learning pipeline to infer travel modes from sparse, unlabeled mobile phone data. First, activities and trips are identified based on spatial and temporal persistence. To address the lack of ground-truth mode labels, the authors use Google Maps API benchmarks to identify a subset of “direct” trips with high-confidence transit or driving labels. These labeled trips train a supervised classifier to distinguish between transit and driving based on features like duration, speed, and distance. A Mahalanobis distance filter separates trips that do not resemble either mode, which are then clustered to identify active travel and other complex modes. Finally, a TI classification model trained on survey data from the Detroit Metro Area Communities Study is transferred to location intelligence data from Chicago, using inferred mode patterns and census-tract-level socio-demographic inputs. The results demonstrate the feasibility of this approach. The semi-supervised framework successfully distinguished driving, transit, and active-mode trips, with aggregate mode shares aligning closely with regional travel surveys. When applied to Chicago data, the transferred TI classification model achieved approximately 70% accuracy and recall, despite simplifying mode categories and using aggregate demographic data instead of individual characteristics. Although direct transfer from Detroit to Chicago introduced some bias, leading to an overestimation of TI, the inferred spatial patterns of insecurity aligned with well-documented areas of mobility disadvantage. The significance of this work lies in establishing a scalable pathway for monitoring transportation equity. By validating that location intelligence data can reliably infer TI status, the study offers a method for systematic, large-scale assessment of transportation access and reliability. This approach enables policymakers to evaluate transportation outcomes and design interventions without the logistical constraints of recurring surveys, although further refinement is needed to mitigate cross-regional bias in transfer learning applications.

Key finding

A transfer learning framework using location intelligence data can infer transportation insecurity with approximately 70% accuracy, producing spatial patterns that align with known areas of mobility disadvantage.

Methodology

modeling

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. Discovered via bulk_ingest_rosap on 2026-05-23 (5 acquisition events logged).

Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.

Topics

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

• exposure measurement

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: observational prevalence