Towards Inclusive Mobility Modeling: Characterizing and Evaluating Elderly Trajectory Patterns in Urban Systems

The Blind Spot in Smart City Data

A new preprint from arXiv (2606.31207) tackles a critical but often overlooked problem in urban mobility AI: the systematic underrepresentation of elderly populations in trajectory datasets. The research proposes methods to characterize and evaluate the movement patterns of older adults, aiming to correct a bias that has quietly shaped—and limited—the development of smart city systems.

What the Research Addresses

The core issue is straightforward yet profound. Most public mobility datasets, whether derived from GPS traces, transit card swipes, or mobile phone pings, are dominated by working-age adults. The elderly, who may travel less frequently, use different modes of transport, or have lower smartphone adoption, generate fewer data points. This sparsity means that AI models trained on these datasets learn to optimize for the majority population while treating elderly mobility patterns as noise or outliers.

The authors propose a framework for explicitly characterizing elderly trajectory patterns—accounting for factors like slower movement speeds, shorter trip distances, more frequent stops, and reliance on paratransit or walking. They then evaluate how well existing mobility models capture these patterns, likely finding significant performance gaps.

Why This Matters Beyond Academic Interest

This is not merely a fairness or inclusion issue—it has concrete, measurable consequences. Smart city applications that rely on mobility models include:

• Public transit optimization: Routes and schedules designed around commuter patterns may leave elderly riders with inadequate service.
• Emergency response planning: Evacuation models that underestimate elderly populations or their slower movement speeds can lead to life-threatening failures.
• Urban infrastructure design: Sidewalk maintenance, bench placement, and crosswalk timing all depend on understanding who moves through a city and how.

When AI systems optimize for the average user, they systematically disadvantage those who deviate from that average. The elderly are not a niche group—they are a rapidly growing demographic in most developed nations.

Implications for AI Practitioners

For those building or deploying mobility models, this research carries several actionable lessons:

• Dataset auditing is insufficient without demographic metadata. Simply knowing that a dataset is "large" or "diverse" in geographic coverage does not guarantee it represents elderly populations proportionally. Practitioners need to actively test for representation gaps.

• Sparsity is not noise. The paper implicitly argues that sparse trajectories from elderly users contain meaningful signal that current models discard. Techniques like few-shot learning, data augmentation, or transfer learning from synthetic elderly trajectories may be necessary.

• Evaluation metrics must be stratified. Reporting overall model accuracy can hide severe underperformance on subgroups. Practitioners should adopt disaggregated evaluation by age, mobility type, and other relevant demographics.

• Data collection strategies need redesign. Passive data collection (e.g., from smartphone apps) will always underrepresent certain groups. Active data collection, partnerships with senior centers, or using alternative data sources (e.g., paratransit records) may be required.

Key Takeaways

• Elderly populations are systematically underrepresented in public mobility datasets, leading to AI models that optimize for working-age adults while ignoring the needs of a growing demographic.
• The research proposes explicit characterization of elderly trajectory patterns (slower speeds, shorter trips, different transport modes) and evaluation of how well existing models capture them.
• For AI practitioners, the key implications are: audit datasets for demographic representation, treat sparse elderly data as signal not noise, stratify evaluation metrics by age group, and reconsider passive data collection strategies.
• Correcting this bias is not just an ethical imperative but a practical one for building smart city systems that serve all citizens effectively.