pith. sign in

arxiv: 2604.16464 · v1 · submitted 2026-04-08 · 📊 stat.AP · cs.LG

Horizon-Aware Forecasting of Passenger Assistance Demand for Rail Station Workforce Planning

Michael Sheehan , Irina Timoshenko This is my paper

Pith reviewed 2026-05-10 17:31 UTC · model grok-4.3

classification 📊 stat.AP cs.LG
keywords passenger assistancedemand forecastingworkforce planningrail stationsProphet modeltime seriesoperational planningrisk framework
0
0 comments X

The pith

A horizon-aware Prophet model cuts rail station forecast errors by up to 76.9% and links to 50% fewer failed passenger assistance deliveries.

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

This paper develops a data-driven system to predict how many passengers will need assistance at individual rail stations and to convert those predictions into practical staff schedules. It trains a horizon-aware Prophet model on multi-source operational records and feeds the outputs into a red-amber-green risk scale that respects service rules and constraints. When tested at LNER-managed stations, the forecasts lowered absolute error by as much as 76.9 percent relative to year-on-year baselines. The same approach coincided with roughly half the usual number of missed assistance requests caused by staff shortages. The full pipeline now runs in production to support daily planning decisions.

Core claim

The authors establish that horizon-aware Prophet forecasts trained on multi-source data achieve substantially higher accuracy than year-on-year baselines for station-level passenger assistance demand. When these forecasts inform workforce plans through an interpretable risk framework, the approach is associated with an approximate 50 percent drop in failed assistance deliveries caused by insufficient staff availability. The system has been deployed in production to guide routine planning at LNER-managed stations.

What carries the argument

A horizon-aware Prophet modelling approach for demand forecasting, paired with a red-amber-green risk framework that converts forecasts into actionable staffing requirements under operational constraints.

If this is right

  • Forecasts enable station managers to set daily staff levels more closely matched to expected needs.
  • The risk framework offers a transparent way to flag days when assistance capacity is likely to be strained.
  • Production use demonstrates the method can be integrated into existing rostering processes.
  • Overall service reliability improves when staffing decisions rest on updated demand predictions rather than historical averages.

Where Pith is reading between the lines

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

  • Similar horizon-aware techniques could apply to forecasting other variable service demands in transport, such as wheelchair requests or lost property volumes.
  • Embedding the red-amber-green scale into staff dashboards might increase adoption by making uncertainty visible without complex statistics.
  • Testing the model on data from different rail networks would show whether the accuracy gains transfer beyond the original operator.

Load-bearing premise

The patterns observed in past passenger assistance requests will continue to hold for future periods even if passenger behavior or external conditions change.

What would settle it

Observing a new period with major schedule changes or external disruptions where the model's absolute error does not stay below the year-on-year baseline by the reported margin, or where forecast-guided staffing shows no reduction in missed assistances.

Figures

Figures reproduced from arXiv: 2604.16464 by Irina Timoshenko, Michael Sheehan.

Figure 1
Figure 1. Figure 1: Station-level passenger assistance requests over the last four years, illustrating strong seasonality and peak demand periods. While assistance bookings can typically be made up to twelve weeks in ad￾vance, booking lead-time behaviour indicates that a substantial proportion of requests are made close to the date of travel [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Cumulative distribution of passenger assistance booking lead times. assistance planning teams review anticipated demand and adjust plans around two weeks before travel, for example through targeted redeployment or over￾time planning. At shorter horizons, for instance a few days before departure, service delivery and station managers are responsible for responding to emerg￾ing demand pressures through react… view at source ↗
Figure 3
Figure 3. Figure 3: Forecast error across horizon buckets for representative stations at hourly (left) and daily (right) resolutions. Solid lines denote asymmetric RMSE and dashed lines denote MAE. forecasts reduce aRMSE relative to the year-on-year baseline by approximately 66% at London Kings Cross, 74% at York, and 66% at Berwick-upon-Tweed at the daily level, with comparable proportional improvements observed at the hourl… view at source ↗
Figure 4
Figure 4. Figure 4: Workforce planning RAG heatmap for York (06:00–21:00, 50 days). Green indicates forecast demand is within primary capacity; Amber indicates secondary capacity is required; Red indicates forecast demand exceeds total capacity. outside the most obvious holiday peak, demonstrating the value of the frame￾work in providing forward visibility of non-trivial high-demand days. Beyond the holiday period, the RAG vi… view at source ↗
Figure 5
Figure 5. Figure 5: Prophet component decomposition for York (Medium II horizon), illustrating the contribution of trend, multiple seasonalities, holiday effects, and external regressors to the final forecast. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Residual diagnostics for York (Medium II horizon), including the residual distri￾bution, Q–Q plot, and residuals over time. Circled observations correspond to periods of engineering works not explicitly captured in the model [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
read the original abstract

Passenger assistance services are essential for accessible rail travel, yet demand varies substantially across stations and over time, creating challenges for workforce planning and staff rostering. This paper presents a data-driven decision support framework for forecasting station-level passenger assistance demand and translating forecasts into workforce plans. The forecasting component applies a horizon-aware Prophet modelling approach using multi-source operational data, while the planning component maps demand forecasts to staffing requirements under service and operational constraints through an interpretable red-amber-green risk framework. The approach has been implemented within a production-grade system to support routine planning and staffing decisions across LNER-managed stations. Results demonstrate improved forecast accuracy relative to year-on-year baseline methods, with absolute error reduced by up to 76.9%, and show that forecast-informed staffing is associated with an approximate 50% reduction in failed passenger assistance deliveries attributable to staff availability. These findings highlight the value of integrating interpretable forecasting with operational work.

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 paper presents a data-driven decision support framework for rail station workforce planning that combines a horizon-aware Prophet model for forecasting station-level passenger assistance demand (using multi-source operational data) with an interpretable red-amber-green risk framework to translate forecasts into staffing requirements under service constraints. The system has been deployed in production for LNER stations. It claims up to 76.9% reduction in absolute forecast error relative to year-on-year baselines and an approximate 50% reduction in failed passenger assistance deliveries associated with the forecast-informed staffing approach.

Significance. If the reported forecasting gains and operational associations are robustly validated, the work demonstrates a practical, production-deployed integration of interpretable time-series forecasting with constrained workforce planning in transportation accessibility services. This addresses a real operational challenge and could inform similar decision-support systems in public transport. The emphasis on horizon-awareness and the red-amber-green mapping are strengths for applied settings, though the paper does not report machine-checked proofs or fully reproducible code artifacts.

major comments (3)
  1. [Abstract and operational evaluation section] Abstract and results on operational impact: the claim that forecast-informed staffing is 'associated with an approximate 50% reduction in failed passenger assistance deliveries attributable to staff availability' is presented without details on the evaluation design (pre/post periods, control stations, difference-in-differences, regression controls for passenger volume changes, or station fixed effects). This makes the attribution vulnerable to confounding by concurrent operational changes, as noted in the observational nature of the assessment.
  2. [Forecasting evaluation and results] Forecasting results: the reported up to 76.9% absolute error reduction versus year-on-year baselines lacks any description of data splits, cross-validation procedure, out-of-sample test set, or statistical significance testing of the improvement. Without these, it is impossible to assess whether the horizon-aware Prophet gains are robust or overfit to the historical data used for both fitting and impact claims.
  3. [Forecasting component description] Methods for horizon-aware Prophet: the implementation details on how the horizon adjustments are incorporated into the Prophet model (e.g., specific modifications to seasonality, changepoints, or uncertainty intervals) and the handling of free parameters (hyperparameters and horizon adjustments) are not sufficiently specified to allow replication or to evaluate sensitivity.
minor comments (2)
  1. [Planning component] The red-amber-green risk framework is described as interpretable but would benefit from an explicit equation or pseudocode showing how demand forecasts are mapped to staffing levels under the service constraints.
  2. [Figures and tables] Figure captions and axis labels in the results should explicitly state the time periods, stations, and error metric (e.g., MAE) used for the 76.9% reduction to improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback, which has helped us identify areas where the manuscript can be strengthened for clarity and rigor. We provide point-by-point responses to the major comments below, indicating revisions where we agree changes are warranted.

read point-by-point responses

  1. Referee: [Abstract and operational evaluation section] Abstract and results on operational impact: the claim that forecast-informed staffing is 'associated with an approximate 50% reduction in failed passenger assistance deliveries attributable to staff availability' is presented without details on the evaluation design (pre/post periods, control stations, difference-in-differences, regression controls for passenger volume changes, or station fixed effects). This makes the attribution vulnerable to confounding by concurrent operational changes, as noted in the observational nature of the assessment.

    Authors: We agree that the operational evaluation section would benefit from expanded methodological detail to better support the reported association. In the revised manuscript, we have added a description of the pre- and post-deployment periods, the regression specification incorporating passenger volume controls and station fixed effects, and an explicit discussion of the observational design's limitations, including the lack of a randomized control group due to the production deployment across all LNER stations. These additions clarify the evaluation approach while acknowledging potential confounding factors. revision: yes

  2. Referee: [Forecasting evaluation and results] Forecasting results: the reported up to 76.9% absolute error reduction versus year-on-year baselines lacks any description of data splits, cross-validation procedure, out-of-sample test set, or statistical significance testing of the improvement. Without these, it is impossible to assess whether the horizon-aware Prophet gains are robust or overfit to the historical data used for both fitting and impact claims.

    Authors: We acknowledge that the forecasting evaluation would be more robust with explicit details on the validation procedure. The revised manuscript now includes a dedicated evaluation subsection describing the time-series data split (using a rolling-origin cross-validation scheme), the out-of-sample test set (final 20% of the temporal range), and paired statistical tests (e.g., Wilcoxon signed-rank) applied to error metrics across stations to assess significance of the improvements. This addresses concerns about potential overfitting and allows readers to evaluate the robustness of the gains. revision: yes

  3. Referee: [Forecasting component description] Methods for horizon-aware Prophet: the implementation details on how the horizon adjustments are incorporated into the Prophet model (e.g., specific modifications to seasonality, changepoints, or uncertainty intervals) and the handling of free parameters (hyperparameters and horizon adjustments) are not sufficiently specified to allow replication or to evaluate sensitivity.

    Authors: We agree that greater specificity on the horizon-aware modifications is needed for replicability. The revised methods section now details how horizon adjustments are implemented: by scaling uncertainty intervals proportionally to the forecast horizon, incorporating horizon-dependent seasonality Fourier terms, and adjusting changepoint priors based on horizon length. We have also added the hyperparameter values, the cross-validation tuning procedure, and pseudocode for the adjustments, enabling sensitivity analysis and replication attempts. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper fits a horizon-aware Prophet model to historical multi-source data and reports accuracy gains versus year-on-year baselines plus an observational association between forecast-informed staffing and fewer failed deliveries. These are standard empirical comparisons and production observations rather than derivations that reduce by construction to the fitted inputs. No equations, self-citations, or ansatzes are shown that make the reported accuracy improvement or 50% association equivalent to the training data by definition. The evaluation steps remain externally falsifiable against baselines and operational records.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The work rests on standard Prophet time-series assumptions and domain-specific operational constraints for staffing; no new entities are postulated. Free parameters are the internal Prophet hyperparameters and any horizon-specific adjustments, which are fitted to the multi-source data.

free parameters (1)
  • Prophet hyperparameters and horizon adjustments
    Standard in Prophet implementations; tuned to fit the rail demand data but not enumerated in the abstract.
axioms (1)
  • domain assumption Historical multi-source operational data contains stable, forecastable patterns of passenger assistance demand
    Implicit in the choice of a fitted time-series model and the claim of improved accuracy over baselines.

pith-pipeline@v0.9.0 · 5453 in / 1422 out tokens · 64043 ms · 2026-05-10T17:31:22.511051+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

15 extracted references · 15 canonical work pages

  1. [1]

    A hybrid approach to time series forecasting: Integrating arima and prophet for im- proved accuracy

    A, S., Christo, M.S., Elizabeth, J.V., 2025. A hybrid approach to time series forecasting: Integrating arima and prophet for im- proved accuracy. Results in Engineering 27, 105703. URL: https: //www.sciencedirect.com/science/article/pii/S2590123025017748, doi:10.1016/j.rineng.2025.105703

    work page doi:10.1016/j.rineng.2025.105703 2025

  2. [2]

    Passenger demand fore- casting in scheduled transportation

    Banerjee, N., Morton, A., Akartunalı, K., 2020. Passenger demand fore- casting in scheduled transportation. European Journal of Operational Re- search 286, 797–810. URL: https://www.sciencedirect.com/science/ article/pii/S0377221719308677, doi: 10.1016/j.ejor.2019.10.032

    work page doi:10.1016/j.ejor.2019.10.032 2020

  3. [3]

    Applying facebook prophet to forecast the passenger flow in a seaport

    Breshanaj, M., Stringa, A., Ramosacaj, M., 2025. Applying facebook prophet to forecast the passenger flow in a seaport. Maritime Technol- ogy and Research URL: https://api.semanticscholar.org/CorpusID: 278956586

    work page 2025

  4. [4]

    Chang, J., Song, X., 2024. A railway passenger flow prediction model based on improved prophet, in: Proceedings of the 2023 4th International Conference on Machine Learning and Computer Application, Association for Computing Machinery, New York, NY, USA. pp. 798–804. URL: https: //doi.org/10.1145/3650215.3650354, doi: 10.1145/3650215.3650354

    work page doi:10.1145/3650215.3650354 2024

  5. [5]

    Forecasting short-term air passenger demand using big data from search engine queries

    Kim, S., Shin, D.H., 2016. Forecasting short-term air passenger demand using big data from search engine queries. Automation in Construction 70, 98–108. URL: https://www.sciencedirect.com/science/article/ pii/S0926580516301303, doi: 10.1016/j.autcon.2016.06.009

    work page doi:10.1016/j.autcon.2016.06.009 2016

  6. [6]

    Predicting customer behavior using prophet algorithm in a real time series dataset

    Liço, L., Enesi, I., Jaiswal, H., 2021. Predicting customer behavior using prophet algorithm in a real time series dataset. European Scientific Journal, ESJ 17, 10. doi: 10.19044/esj.2021.v17n25p10

    work page doi:10.19044/esj.2021.v17n25p10 2021

  7. [7]

    Passenger demand forecasting for railway systems

    Nar, M., Arslankaya, S., 2022. Passenger demand forecasting for railway systems. Open Chemistry 20, 105–119. URL: https://doi.org/10.1515/ chem-2022-0124 , doi: 10.1515/chem-2022-0124 . accessed: 2026-02-25

    work page doi:10.1515/chem-2022-0124 2022

  8. [8]

    Orr report shows satisfaction with assisted rail travel remains unchanged

    Office of Rail and Road, 2024. Orr report shows satisfaction with assisted rail travel remains unchanged. https://www.orr.gov.uk/search-news/ orr-report-shows-satisfaction-assisted-rail-travel-remains-unchanged-it-sets-out-new . Accessed: 2026-02-24. 25

    work page 2024

  9. [9]

    Experiences of Passenger Assist 2024–2025

    Office of Rail and Road, 2025. Experiences of Passenger Assist 2024–2025. Research Report. Office of Rail and Road. London, UK. Produced by M.E.L Research Ltd

    work page 2025

  10. [10]

    Saeed, N., Nguyen, S., Cullinane, K., Gekara, V., Chhetri, P.,

    work page

  11. [11]

    Transport Policy 133, 86–107

    Forecasting container freight rates using the prophet fore- casting method. Transport Policy 133, 86–107. URL: https: //www.sciencedirect.com/science/article/pii/S0967070X23000185, doi:10.1016/j.tranpol.2023.01.012

    work page doi:10.1016/j.tranpol.2023.01.012 2023

  12. [12]

    Four transformations on the Catalan triangle

    Taylor, S., Letham, B., 2018. Forecasting at scale. The American Statisti- cian 72, 37–45. doi: 10.1080/00031305.2017.1380080

    work page Pith review doi:10.1080/00031305.2017.1380080 2018

  13. [13]

    Clustering railway passenger demand patterns from large-scale origin–destination data

    van der Knaap, R.J., de Bruyn, M., van Oort, N., Huisman, D., Goverde, R.M., 2024. Clustering railway passenger demand patterns from large-scale origin–destination data. Journal of Rail Transport Planning & Management 31, 100452. URL: https://www.sciencedirect.com/science/article/ pii/S2210970624000222, doi: 10.1016/j.jrtpm.2024.100452

    work page doi:10.1016/j.jrtpm.2024.100452 2024

  14. [14]

    Passenger Demand Forecasting Handbook

    Worsley, T., 2012. Passenger Demand Forecasting Handbook. Techni- cal Report. RAC Foundation. URL: https://www.racfoundation.org/ wp-content/uploads/2017/11/pdfh-worsley-dec2012.pdf

    work page 2012

  15. [15]

    Systematic review of pas- senger demand forecasting in aviation industry

    Zachariah, R.A., Sharma, S., Kumar, V., 2023. Systematic review of pas- senger demand forecasting in aviation industry. Multimedia Tools and Applications 82, 46483–46519. doi: 10.1007/s11042-023-15552-1 . 26

    work page doi:10.1007/s11042-023-15552-1 2023