pith. sign in

arxiv: 2605.23234 · v1 · pith:AGKHVG7Anew · submitted 2026-05-22 · 💻 cs.LG · cs.CY

Assessing Predictive Models for Fairness Based on Movement Patterns

Francesco Lettich , Mario A. Nascimento , Chiara Pugliese , Chiara Renso This is my paper

Pith reviewed 2026-05-25 05:10 UTC · model grok-4.3

classification 💻 cs.LG cs.CY
keywords spatial fairnessmovement patternspredictive modelsspatial scan statisticfairness assessmentmulti-resolution analysis
0
0 comments X

The pith

Predictive models can be unfair to individuals based on their movement patterns across regions rather than single locations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that spatial fairness assessments must be extended beyond assigning each person to one fixed location like home, because where people actually travel matters for detecting bias in predictions. It introduces a method that links movement data to geographic regions using several map partitions at different scales and alignments, then applies a spatial scan statistic to test for unfair treatment. Experiments on thousands of synthetic datasets demonstrate that this detects the new form of unfairness and identifies the affected individuals. A consistent trade-off appears in how well locations are pinpointed across the different map resolutions.

Core claim

Spatial fairness must be generalized to movement patterns, and an approach that associates movements to regions across multiple spatial partitions with varying resolutions and alignments, then applies a spatial scan statistic, can assess whether a predictive model is fair with respect to those patterns.

What carries the argument

The multi-resolution association of individual movements to geographic regions across different partitions and alignments, followed by a spatial scan statistic that flags unfair treatment.

If this is right

  • The method detects unfairness arising from movement patterns that single-location checks would miss.
  • It identifies the specific set of objects receiving unfair treatment.
  • Localization accuracy shows a consistent trade-off across different spatial resolutions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same movement-based fairness check could be applied to trajectory data from location-based services or transportation systems.
  • Testing the approach on real mobility traces rather than only synthetic cases would reveal how well it handles noisy or incomplete movement records.

Load-bearing premise

Movements of individuals can be meaningfully linked to geographic regions across multiple spatial partitions with different resolutions and alignments so that a spatial scan statistic will correctly identify unfair treatment.

What would settle it

A collection of synthetic datasets containing known movement-based unfairness where the method fails to detect the biased regions or the set of unfairly treated objects.

Figures

Figures reproduced from arXiv: 2605.23234 by Chiara Pugliese, Chiara Renso, Francesco Lettich, Mario A. Nascimento.

Figure 1
Figure 1. Figure 1: shows a heatmap based on a random sample of 1 million trajectory points drawn from the original 720,199,539, in which we notice their uneven spatial distribution. The sampling rate of the trajectories is 2 minutes. The temporal interval spanned by the trajectories is 10 days. The bounding box enclosing the trajectory samples has a width of 4.69 kilometers and a height of 4.83 kilometers. From here on, we r… view at source ↗
Figure 2
Figure 2. Figure 2: Left plot: the subset of 51 US Census block groups intersecting the movement data, shown as red-filled with blue edges. Right plot: hotspot made of two separate regions (blue filled polygons) constructed from the original block groups. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Varying the number of objects per hotspot. dates deemed extreme from all grids exhibits the same trend and maximizes power, thus proving to be a beneficial strategy for unfairness detection. Resolution-wise, the best-performing resolution depends on the number of objects. This suggests that increas￾ing this parameter also tends, on average, to increase hotspot extent, thereby shifting the most suitable res… view at source ↗
Figure 4
Figure 4. Figure 4: Varying the unfairness magnitude. 19 [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Varying the number of hotspots per dataset. best-performing resolutions fall between 400 and 800 meters. Furthermore, observe that as we move from finer to coarser resolutions, the gap between the three configurations tends to narrow. We conjecture that finer grid resolutions tend to fragment hotspots more across grid cells. Consequently, adding more hotspots increases the chance that at least one of them,… view at source ↗
Figure 6
Figure 6. Figure 6: Varying the number of regions per hotspot. 6.4. Number of regions per hotspot We consider the three configurations of 1,000 unfair auditable datasets each, in which we vary the number of regions composing each hotspot. Increasing the number of regions does not increase the number of associated objects, but it requires those same objects to be associated with more distinct regions; as a result, hotspot regi… view at source ↗
Figure 7
Figure 7. Figure 7: Varying the number of stops per hotspot region. observed in the previous experiments. Sensitivity follows similar trends, parameter-wise and resolution-wise. Finally, PPV behaves similarly parameter-wise, while resolution-wise the best performing resolutions are the finer ones, which is in line with the results observed in the previous experiments [PITH_FULL_IMAGE:figures/full_fig_p024_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The plots in the top row show the cells, i.e., the gray-shaded areas, belonging to the extreme candidates detected at coarser grid resolutions. The plots in the bottom row show the cells belonging to candidates detected at finer grid resolutions. The darker an area is, the more it is covered by cells from extreme candidates. The black rectangle represents the bounding box containing the movement data. 25 … view at source ↗
Figure 9
Figure 9. Figure 9: The left plot shows the two actual hotspots, each composed of two regions, highlighted in blue and red, respectively. The yellow and red points show the stop centroids of the objects associated with the hotspots: yellow points fall within the hotspot regions, while red points fall outside them. The right plot superimposes, on the hotspots, the cells belonging to the extreme candidates detected from the 50-… view at source ↗
read the original abstract

Assessing the spatial fairness of predictive models involves establishing whether they are statistically penalizing (favoring) individuals associated with certain geographical locations. Literature on this topic makes the fundamental assumption that each individual is assigned to a single geographical location (e.g., place of residence). However, fairness with respect to the set of locations where one has been, i.e., their movement patterns over different regions, also matters when fairness is considered. Consequently, we argue that it is necessary to generalize the notion of spatial fairness to also include movement patterns, leading to the novel problem of assessing predictive models for fairness relative to the movements of individuals. To deal with this problem, we propose an approach that first associates the movements of individuals to certain geographic regions, considering multiple spatial partitions with different resolutions and alignments, and then employs a suitable spatial scan statistic to assess whether a predictive model is fair based on movement patterns. In the experimental evaluation, we study the performance of our approach over thousands of synthetic unfair datasets, showing that it is effective at detecting this new type of unfairness and at retrieving the set of objects treated unfairly, while localization performance exhibits a consistent multi-resolution trade-off.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper argues that spatial fairness assessment must be generalized beyond single-location assignments to account for individuals' movement patterns across regions. It proposes first mapping movement trajectories to geographic regions over multiple spatial partitions (varying resolutions and alignments), then applying a spatial scan statistic to detect whether a predictive model exhibits unfairness with respect to these patterns. Experiments on thousands of synthetic unfair datasets are reported to show that the approach detects such unfairness, retrieves the set of unfairly treated objects, and exhibits a multi-resolution localization trade-off.

Significance. If the association and detection steps prove reliable, the work would usefully extend spatial fairness auditing to mobility data, which is relevant for applications involving location traces such as ride-sharing or urban services. The multi-partition strategy is a constructive attempt to mitigate alignment sensitivity. No machine-checked proofs, open code, or parameter-free derivations are described.

major comments (2)
  1. [Abstract / Proposed approach] Abstract, paragraph on the proposed approach: the concrete rule for associating individual movement trajectories to regions (visit count, duration, path intersection, or other) across multiple partitions is not specified. This mapping is load-bearing for the central claim, because the input representation fed to the spatial scan statistic depends entirely on it; without an explicit rule, it is impossible to determine whether detected clusters reflect genuine movement-based unfairness or artifacts of the association.
  2. [Abstract / Experimental evaluation] Abstract, experimental evaluation paragraph: no description is given of how the synthetic unfairness was injected into the thousands of datasets, which baselines were used, or whether statistical significance tests were performed. These omissions directly undermine the claim that the method is 'effective at detecting this new type of unfairness and at retrieving the set of objects treated unfairly.'
minor comments (1)
  1. [Abstract] The abstract mentions a 'consistent multi-resolution trade-off' in localization performance but does not define the quantitative measure used to characterize this trade-off.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We address each major comment below and will revise the manuscript to improve clarity.

read point-by-point responses

  1. Referee: [Abstract / Proposed approach] Abstract, paragraph on the proposed approach: the concrete rule for associating individual movement trajectories to regions (visit count, duration, path intersection, or other) across multiple partitions is not specified. This mapping is load-bearing for the central claim, because the input representation fed to the spatial scan statistic depends entirely on it; without an explicit rule, it is impossible to determine whether detected clusters reflect genuine movement-based unfairness or artifacts of the association.

    Authors: We agree that an explicit association rule is necessary. The full manuscript specifies that trajectories are mapped to regions by counting visits within each region across the multiple partitions (with varying resolutions and alignments). To address the comment, we will revise the abstract to state this rule concisely, e.g., 'by associating trajectories to regions via visit counts over multiple spatial partitions'. revision: yes

  2. Referee: [Abstract / Experimental evaluation] Abstract, experimental evaluation paragraph: no description is given of how the synthetic unfairness was injected into the thousands of datasets, which baselines were used, or whether statistical significance tests were performed. These omissions directly undermine the claim that the method is 'effective at detecting this new type of unfairness and at retrieving the set of objects treated unfairly.'

    Authors: We agree the abstract should summarize these elements. The manuscript generates the synthetic datasets by injecting unfairness via biased predictions on targeted movement patterns, uses single-resolution variants as baselines, and relies on the scan statistic's p-value testing for significance. We will revise the abstract's experimental paragraph to include brief descriptions of the injection method, baselines, and significance testing. revision: yes

Circularity Check

0 steps flagged

No circularity; method applies existing spatial scan statistic to new movement-based input representation.

full rationale

The paper introduces a novel problem definition for movement-pattern fairness and proposes an approach that associates trajectories to regions over multiple partitions before applying a spatial scan statistic. No equations, parameter fits, or derivations are shown that reduce any claimed result to its own inputs by construction. Evaluation occurs on independently generated synthetic unfair datasets, providing external verification rather than tautological output. Self-citations, if present for the scan statistic, are not load-bearing for the central claims.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on domain assumptions from spatial statistics and fairness research rather than new free parameters or invented entities.

axioms (2)
  • domain assumption Movement trajectories can be aggregated into region memberships across multiple partitions without losing the signal of unfair treatment.
    Invoked when the method associates movements to geographic regions at different resolutions.
  • domain assumption A spatial scan statistic is an appropriate detector for group-level unfairness defined over movement patterns.
    Central to the detection step described in the abstract.

pith-pipeline@v0.9.0 · 5738 in / 1287 out tokens · 18717 ms · 2026-05-25T05:10:23.291504+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    Proceedings 26th

    Dimitris Sacharidis and Giorgos Giannopoulos and George Papastefanatos and Kostas Stefanidis , title =. Proceedings 26th

    work page

  2. [2]

    Communications in Statistics-Theory and methods , volume=

    A spatial scan statistic , author=. Communications in Statistics-Theory and methods , volume=. 1997 , publisher=

    work page 1997

  3. [3]

    Sina Shaham and Gabriel Ghinita and Cyrus Shahabi , title =. Proc

    work page

  4. [4]

    The Thirty-Fourth

    An Yan and Bill Howe , title =. The Thirty-Fourth. 2020 , url =. doi:10.1609/AAAI.V34I01.5458 , timestamp =

    work page doi:10.1609/aaai.v34i01.5458 2020

  5. [5]

    CoRR , volume =

    Xiaotong Guo and Hanyong Xu and Dingyi Zhuang and Yunhan Zheng and Jinhua Zhao , title =. CoRR , volume =. 2024 , url =. doi:10.48550/ARXIV.2401.00093 , eprinttype =. 2401.00093 , timestamp =

    work page doi:10.48550/arxiv.2401.00093 2024

  6. [6]

    Travel Demand Forecasting: A Fair AI Approach , year=

    Zhang, Xiaojian and Ke, Qian and Zhao, Xilei , journal=. Travel Demand Forecasting: A Fair AI Approach , year=

    work page

  7. [7]

    2023 , eprint=

    Fairness-enhancing deep learning for ride-hailing demand prediction , author=. 2023 , eprint=

    work page 2023

  8. [8]

    2022 , eprint=

    Towards Algorithmic Fairness in Space-Time: Filling in Black Holes , author=. 2022 , eprint=

    work page 2022

  9. [9]

    An Yan and Bill Howe , title =

    work page

  10. [10]

    arXiv preprint arXiv:2403.14040 , year=

    Spatial Fairness: The Case for its Importance, Limitations of Existing Work, and Guidelines for Future Research , author=. arXiv preprint arXiv:2403.14040 , year=

    work page arXiv

  11. [11]

    Proceedings of the 3rd innovations in theoretical computer science conference , pages=

    Fairness through awareness , author=. Proceedings of the 3rd innovations in theoretical computer science conference , pages=

    work page

  12. [12]

    Proceedings of fairness, accountability, transparency in machine learning, 2018 , year=

    Individual fairness under composition , author=. Proceedings of fairness, accountability, transparency in machine learning, 2018 , year=

    work page 2018

  13. [13]

    Proceedings of the Conference on Fairness, Accountability, and Transparency , volume=

    Group fairness under composition , author=. Proceedings of the Conference on Fairness, Accountability, and Transparency , volume=

    work page

  14. [14]

    arXiv preprint arXiv:2201.06598 , year=

    Fairness in federated learning for spatial-temporal applications , author=. arXiv preprint arXiv:2201.06598 , year=

    work page arXiv

  15. [15]

    Adjunct Proceedings of the 2023 ACM International Joint Conference on Pervasive and Ubiquitous Computing & the 2023 ACM International Symposium on Wearable Computing , pages=

    Analysing Fairness of Privacy-Utility Mobility Models , author=. Adjunct Proceedings of the 2023 ACM International Joint Conference on Pervasive and Ubiquitous Computing & the 2023 ACM International Symposium on Wearable Computing , pages=

    work page 2023

  16. [16]

    2015 International Conference on Connected Vehicles and Expo (ICCVE) , pages=

    Privacy preserving driving style recognition , author=. 2015 International Conference on Connected Vehicles and Expo (ICCVE) , pages=. 2015 , organization=

    work page 2015

  17. [17]

    Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency , pages=

    FairAssign: Stochastically Fair Driver Assignment in Gig Delivery Platforms , author=. Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency , pages=

    work page 2023

  18. [18]

    Proceedings of the 1st Conference on Fairness, Accountability and Transparency , pages =

    Runaway Feedback Loops in Predictive Policing , author =. Proceedings of the 1st Conference on Fairness, Accountability and Transparency , pages =. 2018 , volume =

    work page 2018

  19. [19]

    The modifiable areal unit problem (

    Wong, David WS , booktitle=. The modifiable areal unit problem (. 2004 , publisher=

    work page 2004

  20. [20]

    Proceedings of the conference on fairness, accountability, and transparency , pages=

    A comparative study of fairness-enhancing interventions in machine learning , author=. Proceedings of the conference on fairness, accountability, and transparency , pages=

    work page

  21. [21]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Fairness by “Where”: A Statistically-Robust and Model-Agnostic Bi-level Learning Framework , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2022 , month=

    work page 2022

  22. [22]

    Proceedings of the 32nd ACM GIS , pages=

    The Patterns of Life Human Mobility Simulation , author=. Proceedings of the 32nd ACM GIS , pages=

    work page

  23. [23]

    Proceedings of the 31st ACM GIS , pages=

    Massive trajectory data based on patterns of life , author=. Proceedings of the 31st ACM GIS , pages=

    work page

  24. [24]

    Computational and Mathematical Organization Theory , volume=

    Urban life: a model of people and places , author=. Computational and Mathematical Organization Theory , volume=. 2023 , publisher=

    work page 2023

  25. [25]

    International Conference on Geographic Information Science , pages=

    Project Lachesis: parsing and modeling location histories , author=. International Conference on Geographic Information Science , pages=. 2004 , organization=

    work page 2004

  26. [26]

    2008 , issn =

    A conceptual view on trajectories , journal =. 2008 , issn =. doi:https://doi.org/10.1016/j.datak.2007.10.008 , author =

    work page doi:10.1016/j.datak.2007.10.008 2008

  27. [27]

    , author=

    Fast discovery of association rules. , author=. Advances in knowledge discovery and data mining , volume=. 1996 , publisher=

    work page 1996

  28. [28]

    2002 , publisher=

    Scalable algorithms for association mining , author=. 2002 , publisher=

    work page 2002

  29. [29]

    Python for high performance and scientific computing , volume=

    pandas: a foundational Python library for data analysis and statistics , author=. Python for high performance and scientific computing , volume=. 2011 , publisher=

    work page 2011

  30. [30]

    ACM sigmod record , volume=

    Mining frequent patterns without candidate generation , author=. ACM sigmod record , volume=

    work page

  31. [31]

    International journal of health geographics , volume=

    A scan statistic for continuous data based on the normal probability model , author=. International journal of health geographics , volume=. 2009 , publisher=

    work page 2009

  32. [32]

    Statistics in medicine , volume=

    A spatial scan statistic for multinomial data , author=. Statistics in medicine , volume=. 2010 , publisher=

    work page 2010

  33. [33]

    IEEE Data Eng

    GeoLife: A Collaborative Social Networking Service among User, Location and Trajectory , author=. IEEE Data Eng. Bull. , year=

    work page

  34. [34]

    Statistical applications in the spatial sciences , year=

    A million or so correlated coefficients: three experiment on the modifiable areal unit problem , author=. Statistical applications in the spatial sciences , year=

    work page

  35. [35]

    Proceedings of the 28th International Conference on Advances in Geographic Information Systems , pages=

    Evaluating the evaluation metrics for spatial disease cluster detection algorithms , author=. Proceedings of the 28th International Conference on Advances in Geographic Information Systems , pages=

    work page

  36. [36]

    Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection , pages=

    Spatial disease anomaly detection under sparse data with stability metrics , author=. Proceedings of the 1st ACM SIGSPATIAL International Workshop on Geospatial Anomaly Detection , pages=

    work page

  37. [37]

    2015 , issue_date =

    Zheng, Yu , title =. 2015 , issue_date =. doi:10.1145/2743025 , journal =

    work page doi:10.1145/2743025 2015

  38. [38]

    Journal of Statistical Software , volume=

    Scikit-mobility: A Python library for the analysis, generation, and risk assessment of mobility data , author=. Journal of Statistical Software , volume=

    work page

  39. [39]

    Spatial and spatio-temporal epidemiology , volume=

    Measuring the spatial accuracy of the spatial scan statistic , author=. Spatial and spatio-temporal epidemiology , volume=. 2011 , publisher=

    work page 2011

  40. [40]

    arXiv preprint arXiv:2509.23482 , year=

    GeoBS: Information-Theoretic Quantification of Geographic Bias in AI Models , author=. arXiv preprint arXiv:2509.23482 , year=

    work page arXiv

  41. [41]

    , author=

    Legally-Compliant Spatial Fairness Framework: Advancing Beyond Spatial Fairness. , author=. Proceedings 28th EDBT , pages=

    work page

  42. [42]

    International journal of health geographics , volume=

    A nonparametric spatial scan statistic for continuous data , author=. International journal of health geographics , volume=. 2015 , publisher=

    work page 2015

  43. [43]

    International journal of health geographics , volume=

    A flexibly shaped spatial scan statistic for detecting clusters , author=. International journal of health geographics , volume=. 2005 , publisher=

    work page 2005

  44. [44]

    International journal of health geographics , volume=

    Using Gini coefficient to determining optimal cluster reporting sizes for spatial scan statistics , author=. International journal of health geographics , volume=. 2016 , publisher=

    work page 2016

  45. [45]

    Journal of the Royal Statistical Society: Series A (Statistics in Society) , volume=

    Prospective time periodic geographical disease surveillance using a scan statistic , author=. Journal of the Royal Statistical Society: Series A (Statistics in Society) , volume=. 2001 , publisher=

    work page 2001

  46. [46]

    ACM transactions on database systems , volume=

    Schubert, Erich and Sander, J. ACM transactions on database systems , volume=

    work page

  47. [47]

    2025 , howpublished =

    work page 2025

  48. [48]

    ACM Comput

    Caton, Simon and Haas, Christian , title =. ACM Comput. Surv. , articleno =. 2024 , issue_date =

    work page 2024