Learning Individual Movement Shifts After Urban Disruptions with Social Infrastructure Reliance CCS: Information systems Geographic information systems CCS: Applied computing Law, social and behavioral sciences CCS: Computing methodologies Machine learning
Shangde Gao, Zelin Xu, Zhe Jiang · 2025
Read Learning Individual Movement Shifts After Urban Disruptions with Social Infrastructure Reliance CCS: Information systems Geographic information systems CCS: Applied computing Law, social and behavioral sciences CCS: Computing methodologies Machine learning by Shangde Gao, Zelin Xu, Zhe Jiang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-23989.txt.
Introduction
Shangde Gao and two co-authors learn how individuals change their movement after urban disruptions, and how reliance on social infrastructure shapes those shifts. After floods, outages, or closures, people's travel patterns reveal new demands on community resources. Predicting the shifts in advance is hard: resilience is rarely measured, standard features are scarce, and movement data are sparse and tangled with place.
The study builds a measure of each person's social infrastructure reliance and feeds it into a conditioned deep learning model alongside spatial context. Large-scale individual-level data drive the training. Experiments show the added reliance measure and context improve predictions of post-event movement. The model separates people whose pre-event patterns look identical but whose resilience differs — they diverge after disruption, and the model sees it coming. The authors aim the tool at planners who must pre-position community resources before the next shock.
Catalog record
| CREATOR | Shangde Gao, Zelin Xu, Zhe Jiang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 20 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-23989
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Learning Individual Movement Shifts After Urban Disruptions with Social Infrastructure Reliance CCS: Information systems Geographic information systems CCS: Applied computing Law, social and behavioral sciences CCS: Computing methodologies Machine learning Shangde Gao, Zelin Xu, Zhe Jiang Abstract . Shifts in individual movement patterns following disruptive events can reveal changing demands for community resources. However, predicting such shifts before disruptive events remains challenging for several reasons. First, measures are lacking for individuals’ heterogeneous social infrastructure resilience (SIR), which directly influences their movement patterns, and commonly used features are often limited or unavailable at scale, e.g., sociodemographic characteristics. Second, the complex interactions between individual movement patterns and spatial contexts have not been sufficiently captured. Third, individual-level movement may be spatially sparse and not well-suited to traditional decision-making methods for movement predictions. This study incorporates individuals’ SIR into a conditioned deep learning model to capture the complex relationships between individual movement patterns and local spatial context using large-scale, sparse individual-level data. Our experiments demonstrate that incorporating individuals’ SIR and spatial context can enhance the model’s ability to predict post-event individual movement patterns. The conditioned model can capture the divergent shifts in movement patterns among individuals who exhibit similar pre-event patterns but differ in SIR. 1. Introduction Urban dynamics are fundamentally shaped by human movements, which encapsulate how people access daily-life resources ( Gao et al. 2017 ) . Understanding and predicting human movement patterns not only reveal the spatial and temporal rhythms of local communities but also support the development of sustainable and resilient urban environments ( Haraguchi et al. 2022 ) . This becomes critical when communities confront disruptive events—such as natural disasters, pandemics, or infrastructure failures—which may abruptly affect human movements and stress urban systems ( Gong et al. 2025 ) . Predicting movement pattern shifts after disruptive events can help prepare for future event responses by anticipating population needs, guiding emergency response, and strengthening urban adaptability in the post-event normal ( Hong et al. 2021 ) . Figure 1. Illustration of individual movement pattern shifts after urban disruptive events. Previous studies have focused on population-level movement patterns after natural disasters and pandemics ( Lara et al. 2023 ; Gong et al. 2025 ) . While informative for community-level response and adaptation ( Haraguchi et al. 2022 ) , such an approach may obscure individual-level heterogeneity ( Yang et al. 2023 ) . As Figure 1 shows, some people may drastically alter movement patterns after disruptions, while others may adjust minimally ( Zhang and Li 2022 ) . Such heterogeneity can be related to factors of individuals, such as place of residence and surrounding built environment ( Gao et al. 2023 ) , preference in daily activities ( Zhang and Li 2022 ) , and social infrastructure reliance (SIR in short) ( Li and Yan 2024 ) . Accounting for this individual-level heterogeneity can complement population-level analyses and improve urban resilience planning by revealing nuanced behavioral responses, identifying vulnerable groups, and guiding targeted interventions. For example, individuals with poor health conditions may increase visits to health-related social infrastructure, yet these movement shifts can be masked within overall population patterns. Despite its importance, individuals’ movement patterns and shifts after disruptive events remain underexplored due to several key challenges. First, widely-used traits to predict individual movement patterns, such as socio-demographic attributes, may not predict mobility patterns accurately ( Yang et al. 2023 ) and are difficult to obtain at scale due to privacy and cost constraints ( Engin et al. 2020 ) . In contrast, SIR inferred from historical movement patterns ( Zhao et al. 2024 ) can better signal shifts: high-SIR individuals may maintain stable movement patterns, while low-SIR individuals can change movements dynamically ( Zhang and Li 2022 ) . Yet, SIR has been largely overlooked in current models of post-disruption movement patterns. Second, individual movement patterns can be highly context-dependent ( Gong et al. 2025 ) , shaped by the spatial context of the local built environment, such as access to healthcare, transit, and other services ( Gao et al. 2023 ) . Most existing models may not capture the complex interactions between individual movement patterns and spatial context, or they lack the architectural flexibility to integrate spatial context into individual movement pattern predictions. Third, data of individual movement patterns is often sparse in both spatial and temporal perspectives ( Yabe et al. 2022 ) , which may not be well-suited to traditional decision-making methods for movement predictions. This study aims to predict the shifts in individual movement patterns before and after disruptive events, conditioned on individuals’ SIR and the spatial context. SIR is measured as an individual’s likelihood of increasing or decreasing visits to certain social infrastructure when confronting and responding to the disruptions. To address this problem, we employ a conditional UNet model ( Esser et al. 2018 ) , referred to as cUNet. The cUNet model learns the shifts of individuals’ movement patterns from normal periods before disruptive events to the post-event patterns, conditioned on individuals’ SIR across types of social infrastructure and the spatial context of their place of residence. We evaluate the cUNet model using YJMob100K ( Yabe et al. 2024 ) , a large-scale real-world mobility dataset enriched with detailed social infrastructure distribution, aiming to demonstrate the model’s ability to capture nuanced shifts in individual movement patterns and offer a foundation for future advancements in behavior-aware and context-adaptive movement models. In summary, the contributions of this paper are as follows: • Integrating SIR and spatial context into predictions of shifts of individual movement patterns: We incorporate SIR and fine-grained spatial contexts as conditioning inputs, enabling our model to predict nuanced changes in individual-level movement patterns after disruptive events, including divergent behavioral responses among individuals with similar pre-event patterns but differing SIR. This integration can enhance the prediction’s accuracy and interpretability for future behavior-aware urban resilience modeling. • Capturing the complex interactions between individual movement patterns and spatial context: We employ conditioned deep learning models to learn the influence of individual SIR and local spatial context on post-event movement patterns. This approach enables the model to represent heterogeneous behavioral responses conditioned on individual-level and localized traits, which are usually difficult to capture using traditional or population-level models. • Employing a conditional deep learning model to learn from sparse data of individual movement patterns: Our UNet-based architecture with spatial and SIR conditioning can effectively learn individual-level movement pattern changes from limited and sparse individual-level data, enabling robust generalization across diverse populations. 2. Problem Statement Definition 0. Individual Movement Pattern is the spatial footprint of an individual’s activities within a region over a given time period. It captures where an individual has physically visited and serves as a proxy for behavioral routines, spatial preferences, and interactions with the built environment. For an individual u , the pattern is represented by a binary spatial grid \mathbf{V}_{u}\in\{0,1\}^{M\times N} , where v_{u,i,j}=1 if u visited cell (i,j) during the observation period, and v_{u,i,j}=0 otherwise. Let \mathcal{U} denote the set of individuals, each associated with a movement pattern \mathbf{V}_{u} . Definition 0. Social Infrastructure Reliance (SIR) captures an individual’s dependence on social infrastructure (e.g., grocery stores, healthcare, schools, recreation) and potential sensitivity to their availability. It reflects how an individual’s movement