pith. sign in

arxiv: 2605.08801 · v1 · submitted 2026-05-09 · 💻 cs.LG

Data-driven transport modelling without overfit

Peter Vanya , Katar\'ina \v{S}imkov\'a , Rastislav Farka\v{s} This is my paper

Pith reviewed 2026-05-12 02:14 UTC · model grok-4.3

classification 💻 cs.LG
keywords transport modellingtraffic countsdata-driven modelsoverfitting preventionmacroscopic flow modelsexplainable weightsnetwork planning
0
0 comments X

The pith

A protocol builds transport models from traffic counts using explainable weights and controlled complexity to avoid overfitting.

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

The paper seeks to establish that macroscopic transport models, which forecast traffic changes from new roads or closures, can be constructed using only observed traffic counts instead of expensive and biased socio-economic surveys. The method defines an objective function directly on those counts, maintains transparent model weights, and introduces a structured sequence for raising model detail while limiting overfitting. A sympathetic reader would care because this lowers planning costs, improves reliability on real networks, and supports better decisions about infrastructure interventions, as shown through tests on both simple networks and realistic urban cases.

Core claim

The central claim is that an alternative data-driven modelling protocol works with an objective function based on traffic counts, which can be cheaply and reliably obtained; explainable model weights; and a controlled path to increase model complexity and accuracy. The protocol predicts traffic flows after policy interventions and is demonstrated on toy and realistic examples with a suggestion for extension to multimodal systems including public transport.

What carries the argument

The objective function based on traffic counts combined with explainable model weights and a controlled path that raises complexity while preventing overfitting.

If this is right

  • Models can be built without costly population surveys.
  • Weights stay interpretable so planners can understand the results.
  • Accuracy improves in controlled steps without overfitting.
  • Traffic predictions after road changes or closures become feasible with inexpensive data.
  • The same structure applies to multimodal networks with public transport.

Where Pith is reading between the lines

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

  • Planners could update models rapidly using only sensor or camera counts.
  • The controlled-complexity idea might transfer to other flow-prediction tasks such as logistics or utility networks.
  • Direct comparison of predictions against post-intervention counts on real projects would test whether the method generalizes beyond the reported examples.

Load-bearing premise

Traffic counts alone contain enough information to fix model parameters uniquely and the controlled complexity path stops overfitting on realistic networks without extra validation data.

What would settle it

Finding multiple different parameter sets that produce equally low objective values on the same traffic counts, or seeing that added complexity improves training fit but worsens predictions on held-out intervention scenarios.

Figures

Figures reproduced from arXiv: 2605.08801 by Katar\'ina \v{S}imkov\'a, Peter Vanya, Rastislav Farka\v{s}.

Figure 1
Figure 1. Figure 1: FIG. 1: A general scheme of a three-step transport model workflow, which includes trip generation, trip distribution [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: Minimization of the regional model for various [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: A scatter plot of observed vs predicted traffic [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Training and testing error (GEH statistic) as a [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗
read the original abstract

Macroscopic transport modelling aims to predict traffic flows after proposed public policy interventions, such as a new road or railway section or a temporary road closure. As such, it is a vital step in infrastructure planning and development. Traditionally, building a transport model has relied on complex understanding of socio-economic characteristics of the population requiring expensive data collection via surveys, which are prone to biases. Previous numerical frameworks to optimize transport models to fit observed traffic flows are not easily-interpretable and can lead to overfit. We present here an alternative: a data-driven modelling protocol with objective function based on traffic counts, which can be nowadays cheaply and reliably obtained; explainable model weights; and a controlled path to increase model complexity and accuracy. We demonstrate our approach on several toy and realistic examples, and suggest ways to generalize to multimodal systems including public transport.

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

3 major / 2 minor

Summary. The manuscript proposes a data-driven protocol for macroscopic transport modeling as an alternative to traditional survey-based approaches. The protocol defines an objective function based solely on observed traffic counts, incorporates explainable model weights, and provides a controlled mechanism for increasing model complexity to achieve higher accuracy without overfitting. Demonstrations are provided on toy networks and realistic examples, with suggestions for extension to multimodal systems including public transport.

Significance. If the central claims hold, the work could lower barriers to building and updating transport models by replacing expensive, biased surveys with routinely collected traffic counts while maintaining interpretability and stability under policy interventions. The controlled-complexity path and emphasis on explainable weights are potentially valuable contributions in a field where black-box optimization often leads to unstable predictions. However, the significance is limited by the absence of explicit handling of parameter identifiability and out-of-sample validation on counterfactual scenarios.

major comments (3)
  1. [§2] §2 (Proposed Protocol): The objective function is stated to depend only on traffic counts, yet the manuscript provides no analysis of the rank deficiency inherent in the link-count mapping. In standard assignment models the count vector lies in a lower-dimensional subspace than the parameter space (OD matrix or route flows); without explicit regularization, null-space components, or uniqueness proofs, multiple parameter sets can achieve identical objective values while producing divergent predictions after interventions such as new links or closures. This directly undermines the 'without overfit' guarantee.
  2. [§3.2] §3.2 (Realistic examples): The reported fits to observed counts are shown, but no quantitative out-of-sample evaluation on policy interventions is presented. In-sample reproduction of counts does not establish that the controlled-complexity path yields stable forecasts under network changes; the demonstrations therefore do not yet substantiate the central claim that the protocol avoids overfitting for its intended use case.
  3. [§2.3] §2.3 (Complexity control): The mechanism for 'controlled path to increase model complexity' is described at a high level but lacks a formal definition (e.g., no equation specifying the penalty term, cross-validation procedure, or information criterion). Without this, it is impossible to verify that the path enforces uniqueness or prevents the trade-off of one form of under-determination for another.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'explainable model weights' is used without contrasting it to the 'not easily-interpretable' previous frameworks; a brief clarification of the interpretability criterion would help readers.
  2. [§4] §4 (Generalization): The multimodal extension is sketched in one paragraph; adding even a small illustrative example or pseudocode would make the claim more concrete.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and insightful comments, which help strengthen the manuscript's claims regarding identifiability, validation, and formalization of the protocol. We address each major comment below and will revise the manuscript to incorporate the suggested improvements where feasible.

read point-by-point responses

  1. Referee: The objective function is stated to depend only on traffic counts, yet the manuscript provides no analysis of the rank deficiency inherent in the link-count mapping. In standard assignment models the count vector lies in a lower-dimensional subspace than the parameter space; without explicit regularization, null-space components, or uniqueness proofs, multiple parameter sets can achieve identical objective values while producing divergent predictions after interventions.

    Authors: We acknowledge that the manuscript lacks an explicit analysis of rank deficiency and null-space components in the link-count mapping. The explainable weights are designed to promote interpretability and constrain solutions toward physically meaningful parameters, but this does not constitute a formal uniqueness proof. In the revision we will add a subsection in §2 discussing the linear dependence structure of the assignment operator and illustrating, via a small analytical example on a toy network, how the controlled-complexity path combined with weight explainability selects stable solutions that remain consistent under interventions such as link additions or closures. revision: partial

  2. Referee: The reported fits to observed counts are shown, but no quantitative out-of-sample evaluation on policy interventions is presented. In-sample reproduction of counts does not establish that the controlled-complexity path yields stable forecasts under network changes; the demonstrations therefore do not yet substantiate the central claim that the protocol avoids overfitting for its intended use case.

    Authors: The current examples focus on in-sample reproduction to demonstrate the fitting procedure and complexity control. We agree that quantitative out-of-sample evaluation on counterfactual policy scenarios is necessary to fully support the 'without overfit' claim. In the revised §3.2 we will add synthetic experiments on both toy and realistic networks: we will simulate interventions (new links, closures), hold out post-intervention counts, and report predictive error metrics for models along the controlled-complexity path versus unconstrained baselines. revision: yes

  3. Referee: The mechanism for 'controlled path to increase model complexity' is described at a high level but lacks a formal definition (e.g., no equation specifying the penalty term, cross-validation procedure, or information criterion). Without this, it is impossible to verify that the path enforces uniqueness or prevents the trade-off of one form of under-determination for another.

    Authors: We will revise §2.3 to supply the missing formal definition. The updated text will include the explicit penalty term added to the objective function, the precise rule for incrementing complexity (based on an information criterion applied to held-out traffic counts), and a brief proof sketch showing that each increment reduces the effective degrees of freedom while preserving consistency with observed data. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The abstract describes a data-driven protocol whose objective is defined directly from observed traffic counts, with explainable weights and a controlled complexity path. No equations, parameter-fitting steps, or self-citations are supplied that would reduce any claimed prediction or uniqueness result to the input data by construction. The demonstrations on toy and realistic examples are presented as external validation rather than tautological re-statements of the fit. Because no load-bearing step can be quoted that exhibits self-definition, fitted-input renaming, or imported uniqueness from the authors' prior work, the derivation does not collapse to its inputs. This is the normal, non-circular outcome for a protocol whose central mechanism (count-based objective plus complexity control) is stated independently of the target performance claims.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; the protocol is described at a high level only.

pith-pipeline@v0.9.0 · 5444 in / 976 out tokens · 42442 ms · 2026-05-12T02:14:58.522970+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

29 extracted references · 29 canonical work pages

  1. [1]

    C. R. Sunstein,The Cost-Benefit Revolution(MIT Press, Cambridge, MA, 2018)

    work page 2018

  2. [2]

    Treiber and A

    M. Treiber and A. Kesting,Traffic flow dynamics: data, models and simulation(Springer, 2013)

    work page 2013

  3. [3]

    Helbing and P

    D. Helbing and P. Moln´ ar, Phys. Rev. E51, 4282 (1995)

    work page 1995

  4. [4]

    A. G. Wilson,Urban and Regional Models in Geography and Planning(John Wiley & Sons, Ltd, Chichester, UK, 1974)

    work page 1974

  5. [5]

    S. C. Dafermos and F. T. Sparrow, Journal of Research of the National Bureau of Standards, Section B: Mathe- matical Sciences73B, 91 (1969)

    work page 1969

  6. [6]

    de Dios Ort´ uzar and L

    J. de Dios Ort´ uzar and L. G. Willumsen,Modelling Transport, 4th ed. (Wiley, 2011)

    work page 2011

  7. [7]

    C. M. Bishop,Pattern Recognition and Machine Learn- ing(Springer, New York, 2006)

    work page 2006

  8. [8]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton (2012) pp. 84 – 90

    work page 2012

  9. [9]

    H. J. van Zuylen and L. G. Willumsen, Transportation Research Part B: Methodological14, 281 (1980)

    work page 1980

  10. [10]

    L. G. Willumsen, (1978)

    work page 1978

  11. [11]

    L. G. Willumsen, inProceedings of the PTRC Sum- mer Annual Meeting 1978, Seminar in Transport Models (PTRC Education and Research Services Ltd., London, UK, 1978)

    work page 1978

  12. [12]

    L. G. Willumsen, Transportation10, 257 (1981). [14]Digitale Plattform Stadtverkehr – Berlin(2026)

    work page 1981

  13. [13]

    D. B. Lee, L. A. Klein, and G. Camus, Journal of the Transportation Research Board1659, 68 (1999)

    work page 1999

  14. [14]

    Maximum effective speed is lower not the same as legal speed, since it is influenced by road width, slope, curva- ture and the level of surface degradation

    work page

  15. [15]

    Bureau of Public Roads,Traffic Assignment Manual, Ur- ban Planning Division, US Department of Commerce, Washington, DC (1964)

    work page 1964

  16. [16]

    As a rule of thumb,α 1 = 0.15 andα 2 = 4 are reasonable estimates

    work page

  17. [17]

    E. T. Jaynes, Physical Review106, 620 (1957)

    work page 1957

  18. [18]

    A. G. Wilson,Entropy in Urban and Regional Modelling (Pion, London, UK, 1970)

    work page 1970

  19. [19]

    Compared to the general macroscopic transport- modelling framework as described in the Introduction, the step of mode choice is missing, as this framework is focused on private-car traffic for now

    work page

  20. [20]

    To convert between persons to passenger vehicles on the road, person-trips must be divided by average vehicle oc- cupancy, typically between 1 and 2 persons/veh

    work page

  21. [21]

    J. Tang, L. Wan, J. Schooling, P. Zhao, J. Chen, and S. Wei, Cities129, 103833 (2022)

    work page 2022

  22. [22]

    H. Song, H. Liang, H. Li, Z. Dai, and X. Yun, European Transport Research Review11(2019), 10.1186/s12544- 019-0390-4

    work page doi:10.1186/s12544- 2019

  23. [23]

    Kirkpatrick, C

    S. Kirkpatrick, C. D. Gelatt, and M. P. Vecchi, Science 220, 671 (1983)

    work page 1983

  24. [24]

    Xiang, D

    Y. Xiang, D. Y. Sun, W. Fan, and X. G. Gong, Physics Letters A233, 216 (1997)

    work page 1997

  25. [25]

    J. A. Nelder and R. Mead, The Computer Journal7, 308 (1965)

    work page 1965

  26. [26]

    [30]https://ec.europa.eu/eurostat/web/nuts

    Available athttps://github.com/petervanya/ traffic-flow. [30]https://ec.europa.eu/eurostat/web/nuts. [31]https://www.ssc.sk/sk/cinnosti/ rozvoj-cestnej-siete/dopravne-inzinierstvo/ celostatne-scitanie-dopravy-v-roku-2015.ssc

    work page 2015

  27. [27]

    ˇSul´ ıkov´ a, P

    S. ˇSul´ ıkov´ a, P. Vanya, L. Kov´ aˇ c, D. Buc, and R. Farkaˇ s, Transportation Research Procedia78, 546 (2024)

    work page 2024

  28. [28]

    M. E. Ben-Akiva and S. R. Lerman,Discrete Choice Analysis: Theory and Application to Travel Demand (MIT Press, Cambridge, Massachusetts, 1985)

    work page 1985

  29. [29]

    M. E. Ben-Akiva and M. Bierlaire, inThe Handbook of Transportation Science, edited by R. G. Hall (Kluwer Academic Publishers, Dordrecht, Netherlands, 1999) pp. 5–33

    work page 1999