arxiv: 2605.13742 · v1 · submitted 2026-05-13 · 📊 stat.ME · stat.AP

Recognition: no theorem link

Macroscopic Activity-Based Modeling of Urban Active Mobility

Adrien Marion, Florian Patout, Romain Aza\"is

Pith reviewed 2026-05-14 17:38 UTC · model grok-4.3

classification 📊 stat.ME stat.AP

keywords activity-based modelingurban mobilityPoisson likelihoodmaximum likelihood estimationEM algorithmsensor data disaggregationprivacy-preserving inferencestatistical mobility modeling

0 comments

The pith

A macroscopic model infers urban traveler subpopulation sizes from aggregated sensor counts via attendance functions and Poisson maximum likelihood.

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

The paper builds a city-scale model of walking and cycling trips that starts from summed counts recorded by non-intrusive sensors rather than individual trajectories. It defines attendance functions that encode the typical times and places people move between daily activities such as home, work, or shops. These functions turn the disaggregation task into a statistical estimation problem in which observed counts are treated as Poisson random variables whose parameters depend on unknown group sizes. Maximum-likelihood estimates of those sizes are obtained with an EM algorithm that carries theoretical convergence guarantees. Because the approach works entirely with aggregates, it remains computationally tractable for large cities while avoiding the privacy issues of trajectory data.

Core claim

Grounded in an underlying microscopic stochastic process, the framework shows that attendance functions can be used to express the expected flow between activity pairs at each time step; the resulting Poisson likelihood then yields consistent maximum-likelihood estimators for the latent subpopulation sizes, which can be computed efficiently by an expectation-maximization procedure.

What carries the argument

Attendance functions that map each pair of activity locations and each time interval to a probability distribution over traveler groups, thereby parameterizing the Poisson means for the observed aggregate counts.

If this is right

Subpopulation sizes are recoverable from summed counts alone, without storing or processing individual movement records.
The EM algorithm supplies both point estimates and a practical way to scale the computation to city-wide sensor networks.
Theoretical consistency results guarantee that the estimates converge to the true sizes as the number of observation periods grows.
The same attendance-function representation can be reused to simulate future mobility scenarios by changing the underlying activity schedule.
Only aggregate data are required, so the method can be applied to existing loop counters or Bluetooth sensors without new privacy agreements.

Where Pith is reading between the lines

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

Cities could use the inferred flows to test the impact of new bike lanes or pedestrian zones before construction by simply altering the attendance parameters.
If attendance functions are learned from one city, they might transfer to another city with similar activity patterns, offering a low-data starting point for new deployments.
Relaxing the Poisson assumption to allow overdispersion would be a direct next step if real count variance exceeds the model prediction.
Coupling the model with land-use data could let planners predict how changes in shop opening hours alter overall active-mobility demand.

Load-bearing premise

The newly defined attendance functions must correctly describe the real probabilities that different groups travel between activities at given times.

What would settle it

Run the estimator on real sensor counts and check whether the recovered subpopulation sizes produce flow predictions that systematically mismatch independent manual counts or travel surveys collected in the same city and period.

Figures

Figures reproduced from arXiv: 2605.13742 by Adrien Marion, Florian Patout, Romain Aza\"is.

**Figure 2.** Figure 2: Spatial distribution of activity locations (left) and distribution of end time of each activity (right). [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Generated subpopulation sizes for the first [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

**Figure 4.** Figure 4: Hidden data for different days, at a single counter location [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗

**Figure 5.** Figure 5: (Empirical approximation of the) attendance function at location [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗

**Figure 6.** Figure 6: (Empirical approximation of the) attendance as a function of the counter location for each journey [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Convergence of our EM algorithm toward the MLE (left) and absolute difference in log-scale [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Boxplots of the MLE for the four components of [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 9.** Figure 9: The six different 2D slices of the 4D confidence ellipsoids at [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗

**Figure 10.** Figure 10: Three distribution strategies (in different colors) for positioning the counters over the spatial [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗

**Figure 11.** Figure 11: Random trajectory of an individual p. It begins at x 0 p , drawn from the red distribution (doors) and arrives at x e p , drawn from the blue distribution (buildings). The green distribution, used to draw the starting time t 0 p , depends on x 0 p . The velocity vp is also drawn from a distribution fv (not shown in the graphic). At time t P Tp, the individual is located at γpptq. the product Nk şti ti´1 ν… view at source ↗

**Figure 12.** Figure 12: (Theoretical) attendance at location x “ 0.53 for each journey type and each hour of the day, computed from formulas (21), (18) and (19) and spatial and temporal distributions of [PITH_FULL_IMAGE:figures/full_fig_p017_12.png] view at source ↗

**Figure 13.** Figure 13: (Theoretical) attendance as a function of the counter location for each journey type and some hour [PITH_FULL_IMAGE:figures/full_fig_p018_13.png] view at source ↗

read the original abstract

This paper develops a macroscopic, activity-based model of urban active mobility using nonintrusive sensor data. It introduces attendance functions to describe spatio-temporal travel patterns between activities and formulates the disaggregation of aggregated counts as a statistical inference problem. Counts are modeled as Poisson variables, and unknown subpopulation sizes are estimated via maximum likelihood, with theoretical guarantees and an efficient EM algorithm for computation. Grounded in a microscopic stochastic model, the framework offers a scalable and privacy-preserving approach to analyzing urban soft mobility dynamics.

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.

Desk Editor's Note private letter to a colleague

The paper introduces attendance functions and frames disaggregating sensor counts as a Poisson MLE problem solved by EM, grounded in a micro model.

read the letter

Hi colleague, The main thing to know about this paper is that it proposes attendance functions to model travel patterns between activities and then uses a Poisson-based maximum likelihood approach with an EM algorithm to estimate subpopulation sizes from aggregated sensor counts. This is presented as a macroscopic model grounded in a microscopic stochastic process. What stands out is the integration of these elements into a scalable, privacy-friendly framework for city-scale active mobility analysis. The use of EM for computation is practical, and the claim of theoretical guarantees adds some weight if the conditions are properly derived. On the downside, the central estimation relies on the counts being independent Poissons with means given by the attendance functions times subpopulation sizes. Real mobility data tends to have dependencies across time and space, which could make the estimator inconsistent. The paper would need to address how it handles potential misspecification of the attendance functions or overdispersion. Since the full validation isn't detailed in what I've seen, it's hard to gauge how well it performs on actual data. This work is aimed at researchers in urban mobility modeling and statistical inference for transportation. Someone working on activity-based models or disaggregation techniques could pick up useful ideas here. Overall, it looks like it should go to peer review so the math and assumptions can be scrutinized properly.

Referee Report

2 major / 1 minor

Summary. The paper develops a macroscopic activity-based model for urban active mobility from non-intrusive sensor data. It introduces attendance functions to capture spatio-temporal travel patterns between activities, models aggregated counts as independent Poisson random variables whose means are linear combinations of unknown subpopulation sizes weighted by these functions, and recovers the subpopulation sizes by maximum-likelihood estimation implemented via an efficient EM algorithm. The framework is stated to be grounded in a separate microscopic stochastic model and to provide theoretical guarantees while remaining scalable and privacy-preserving.

Significance. If the Poisson independence assumption and correct specification of the attendance functions hold, the approach would supply a scalable, privacy-preserving route to disaggregating sensor counts into activity-based mobility flows, with the microscopic grounding and EM algorithm constituting concrete strengths for practical deployment in urban planning.

major comments (2)

[Abstract] Abstract: the claim of 'theoretical guarantees' for the MLE is not accompanied by any statement of regularity conditions, identifiability results, or consistency proofs; without these it is impossible to verify whether the estimator remains consistent when the Poisson independence assumption is violated by temporal autocorrelation or spatial dependence typical in mobility data.
[Estimation procedure] Estimation procedure (central inference step): the MLE consistency and the claimed guarantees rest on the attendance functions being correctly specified so that the mean structure matches the true data-generating process; no validation experiments, sensitivity checks, or comparison against the underlying microscopic model are described to confirm this, leaving the recovery of subpopulation sizes vulnerable to bias.

minor comments (1)

[Abstract] The abstract would benefit from a brief sentence clarifying the precise scope of the theoretical guarantees (e.g., consistency under what conditions on the attendance functions).

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments, which help clarify the scope of our theoretical claims and the need for additional validation. We address each major comment below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: the claim of 'theoretical guarantees' for the MLE is not accompanied by any statement of regularity conditions, identifiability results, or consistency proofs; without these it is impossible to verify whether the estimator remains consistent when the Poisson independence assumption is violated by temporal autocorrelation or spatial dependence typical in mobility data.

Authors: We agree that the abstract is too terse on this point. The guarantees refer to consistency and asymptotic normality of the MLE under the Poisson model with correctly specified attendance functions and the standard regularity conditions for exponential-family MLEs (detailed in Section 3.2). We will revise the abstract to state these conditions explicitly and add a short paragraph in Section 3 discussing robustness to mild dependence; new simulation results assessing sensitivity to temporal autocorrelation will be included. revision: yes
Referee: [Estimation procedure] Estimation procedure (central inference step): the MLE consistency and the claimed guarantees rest on the attendance functions being correctly specified so that the mean structure matches the true data-generating process; no validation experiments, sensitivity checks, or comparison against the underlying microscopic model are described to confirm this, leaving the recovery of subpopulation sizes vulnerable to bias.

Authors: The referee correctly identifies that consistency requires correct specification. Section 2 derives the macroscopic mean structure as the exact expectation of the microscopic process, but we acknowledge the absence of direct numerical validation. We will add a dedicated subsection with Monte Carlo experiments that simulate data from the microscopic model, recover subpopulation sizes via the EM algorithm, and report bias/variance under both correct and mildly misspecified attendance functions. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained via external microscopic grounding and standard inference

full rationale

The paper explicitly grounds the macroscopic activity-based model in a separate microscopic stochastic model, supplying an independent anchor for the attendance functions and overall framework. The core step formulates disaggregation of aggregated sensor counts as a Poisson MLE problem for subpopulation sizes, which is the stated inference objective rather than a derived prediction. No equation or claim reduces by construction to its own inputs; the EM algorithm and theoretical guarantees operate under the model's stated assumptions without self-referential fitting loops or load-bearing self-citations. This is the normal case of a statistical modeling paper whose central claim (recovering sizes from counts) is not equivalent to the data by definition.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The central claim rests on the Poisson count assumption and the existence of attendance functions that are introduced without external validation; subpopulation sizes are treated as free parameters recovered from data.

free parameters (1)

subpopulation sizes
Unknown group sizes are estimated by maximum likelihood from the aggregated sensor counts.

axioms (1)

domain assumption Travel counts between activities follow a Poisson distribution
Standard modeling choice for non-negative integer counts but invoked without justification or robustness checks in the abstract.

invented entities (1)

attendance functions no independent evidence
purpose: Describe spatio-temporal patterns of presence at activities to enable disaggregation of counts
Newly defined objects in the paper with no independent evidence supplied in the abstract.

pith-pipeline@v0.9.0 · 5369 in / 1349 out tokens · 42048 ms · 2026-05-14T17:38:45.617260+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

38 extracted references · 1 canonical work pages

[1]

2019 , publisher=

Akande, Adeoluwa and Cabral, Pedro and Gomes, Paulo and Casteleyn, Sven , journal=. 2019 , publisher=

2019
[2]

2008 , publisher=

Banister, David , journal=. 2008 , publisher=

2008
[3]

2020 , publisher=

Barbarossa, Luca , journal=. 2020 , publisher=

2020
[4]

2015 , publisher=

Bonnel, Patrick and Hombourger, Etienne and Olteanu-Raimond, Ana-Maria and Smoreda, Zbigniew , journal=. 2015 , publisher=

2015
[5]

2013 , publisher=

Chandra, Satish and Bharti, Anish Kumar , journal=. 2013 , publisher=

2013
[6]

Chenu, Alain , journal=
[7]

1977 , publisher=

Dempster, Arthur P and Laird, Nan M and Rubin, Donald B , journal=. 1977 , publisher=

1977
[8]

2014 , publisher=

Depeau, Sandrine and Quesseveur, Erwan , journal=. 2014 , publisher=

2014
[9]

2019 , publisher=

Fekih, Mariem and Bonnel, Patrick and Smoreda, Zbigniew and Bellemans, Tom and Furno, Angelo and Galland, Stéphane , journal=. 2019 , publisher=

2019
[10]

2025 , publisher=

Ganault, Jeanne and Rauch, Capucine , journal=. 2025 , publisher=

2025
[11]

2016 , publisher=

Gebhardt, Laura and Krajzewicz, Daniel and Oostendorp, Rebekka and Goletz, Mirko and Greger, Konstantin and Klötzke, Matthias and Wagner, Peter and Heinrichs, Dirk , journal=. 2016 , publisher=

2016
[12]

2023 , publisher=

Keusch, Florian and Bähr, Sebastian and Haas, Georg-Christoph and Kreuter, Frauke and Trappmann, Mark , journal=. 2023 , publisher=

2023
[13]

2025 , publisher=

Liu, Ranqi and Liu, Ning , journal=. 2025 , publisher=

2025
[14]

2021 , publisher=

Livingston, Mark and McArthur, David and Hong, Jinhyun and English, Kirstie , journal=. 2021 , publisher=

2021
[15]

doi:doi:10.5281/zenodo.11111161 , year=

Vallée, J and Douet, A and Le Roux, G and Commenges, H and Lecomte, C and Villard, E , title=. doi:doi:10.5281/zenodo.11111161 , year=

work page doi:10.5281/zenodo.11111161
[16]

McLachlan, Geoffrey J and Krishnan, Thriyambakam , year=
[17]

2007 , publisher=

Murray, James D , volume=. 2007 , publisher=

2007
[18]

Rabaud, Mathieu , booktitle=
[19]

2018 , month=

Raballand, Wilfried and Le Corre, Maxime , url=. 2018 , month=

2018
[20]

2018 , publisher=

Rodriguez-Carrion, Alicia and Garcia-Rubio, Carlos and Campo, Celeste , journal=. 2018 , publisher=

2018
[21]

2016 , publisher=

Siła-Nowicka, Katarzyna and Vandrol, Jan and Oshan, Taylor and Long, Jed A and Demšar, Urška and Fotheringham, A Stewart , journal=. 2016 , publisher=

2016
[22]

2000 , publisher=

Van der Vaart, Aad W , volume=. 2000 , publisher=

2000
[23]

2015 , publisher=

Xu, Yang and Shaw, Shih-Lung and Zhao, Ziliang and Yin, Ling and Fang, Zhixiang and Li, Qingquan , journal=. 2015 , publisher=

2015
[24]

Yang, Hong and Ozbay, Kaan and Bartin, Bekir , booktitle=
[25]

Les enqu

Cirillo, Cinzia and Corn. Les enqu. Reflets et perspectives de la vie. 2004 , volume =

2004
[26]

, title =

Colomb, M. , title =. NetMob 2025 -- 9th edition of NetMob , year =

2025
[27]

and Perret, J

Colomb, M. and Perret, J. , title =
[28]

ESAIM: Proceedings and Surveys , volume=

Crowd motion and evolution PDEs under density constraints , author=. ESAIM: Proceedings and Surveys , volume=. 2018 , publisher=

2018
[29]

Mathematical Models and Methods in Applied Sciences , volume=

A macroscopic crowd motion model of gradient flow type , author=. Mathematical Models and Methods in Applied Sciences , volume=. 2010 , publisher=

2010
[30]

Traffic and Granular Flow’07 , pages=

Handling of contacts in crowd motion simulations , author=. Traffic and Granular Flow’07 , pages=. 2009 , publisher=

2009
[31]

Transportation Research Part B: Methodological , volume=

A continuum theory for the flow of pedestrians , author=. Transportation Research Part B: Methodological , volume=. 2002 , publisher=

2002
[32]

Annual review of fluid mechanics , volume=

The flow of human crowds , author=. Annual review of fluid mechanics , volume=. 2003 , publisher=

2003
[33]

Physical review E , volume=

Social force model for pedestrian dynamics , author=. Physical review E , volume=. 1995 , publisher=

1995
[34]

nature , volume=

The statistics of crowd fluids , author=. nature , volume=. 1971 , publisher=

1971
[35]

Crowd Dynamics, Volume 4: Analytics and Human Factors in Crowd Modeling , pages=

Time-continuous microscopic pedestrian models: an overview , author=. Crowd Dynamics, Volume 4: Analytics and Human Factors in Crowd Modeling , pages=. 2023 , publisher=

2023
[36]

Active Particles, Volume 1: Advances in Theory, Models, and Applications , pages=

Follow-the-leader approximations of macroscopic models for vehicular and pedestrian flows , author=. Active Particles, Volume 1: Advances in Theory, Models, and Applications , pages=. 2017 , publisher=

2017
[37]

A theory of traffic flow on long crowded roads , author=

On kinematic waves II. A theory of traffic flow on long crowded roads , author=. Proceedings of the royal society of london. series a. mathematical and physical sciences , volume=. 1955 , publisher=

1955
[38]

Operations research , volume=

Shock waves on the highway , author=. Operations research , volume=. 1956 , publisher=

1956